DEV Community: George Anadiotis

Beyond Context Graphs: How Ontology, Semantics, and Knowledge Graphs Define Context. The Year of the Graph Newsletter Vol 30

George Anadiotis — Thu, 19 Mar 2026 09:32:05 +0000

What are context graphs, what are they good for, and why are they dubbed AI’s trillion-dollar opportunity? What does context mean actually, and how can we define context using graphs and ontologies? And how can different types of graphs and graph technologies power AI?

By George Anadiotis

Gartner highlighted Data Management, Semantic Layers, and GraphRAG as Top Trends in Data and Analytics for 2026. Startups and incumbents in the graph technology space are making progress, while graph is becoming the fastest growing segment in AI research.

A comprehensive, up-to-date repository, visualization, and analysis of offerings across the graph technology space has been unveiled. New and existing combinations of Graphs and AI are being used to power use cases such as software engineering productivity and supporting enterprise needs at Netflix scale.

New graph database products, features, and benchmarks are available. Use cases as well as research and development on ontologies are on the rise too, including topics such as Enterprise Architecture, visual tooling, and quality assessment for LLM-assisted use of ontologies.

And yet, the most widely discussed topic in the world of graph technology – and beyond – for this past couple of months has been context graphs. So what are context graphs and where do they fit in the graph technology landscape?

In this issue of the Year of the Graph, we explore progress in Ontology, Semantics, Knowledge Graphs, Graph Databases and Analytics, and how these technologies can help define context and power AI.

An Introduction to Context Graphs
Context Beyond Context Graphs
Ontologies, Context Graphs, and Semantic Layers: What AI Actually Needs in 2026
Tooling and Evaluation Frameworks for Ontologies
From Retrieval Augmented Generation to Knowledge Augmented Generation
Knowledge Graphs in Software Engineering and Enterprise Architecture
Knowledge Graph Research, Applications and Best Practices
Knowledge Graph Tools and Platforms
The State of the Graph Database Market
Graph Analytics and Graph AI Updates

This issue of the Year of the Graph is brought to you by metaphacts, Graphwise, Connected Thinking, Linkurious, Process Tempo, State of the Graph, Connected Data London, and Pragmatic AI Training.

If you want to be featured in an upcoming issue and support this work, reach out!

Why Most Enterprise AI Strategies Fail

Even as AI adoption soars, 80% of enterprises report no return on their AI investments, and 42% end up abandoning their strategies entirely. At the same time, those who pivot away from AI risk accelerating their own obsolescence. The missing link? A knowledge graph with a semantic layer.

By pairing LLMs with a symbolic layer, companies are able to leverage AI and trust that its outputs are contextualized and explainable. This whitepaper dives into how

knowledge graphs provide the necessary structure and grounding that LLMs lack, enabling scalable, future-proof AI strategies.

Download the free whitepaper

An Introduction to Context Graphs

Rules tell an agent what should happen in general. Decision traces capture what happened in specific cases. Agents don’t just need rules. They need access to the decision traces that show how rules were applied in the past, where exceptions were granted, how conflicts were resolved, who approved what, and which precedents actually govern reality.

A context graph is the accumulated structure formed by those traces: not “the model’s chain-of-thought,” but a living record of decision traces stitched across entities and time so precedent becomes searchable. Over time, that context graph becomes the real source of truth for autonomy – because it explains not just what happened, but why it was allowed to happen.

This is how Foundation Capital’s Jaya Gupta and Ashu Garg defined context graphs, claiming they will be the single most valuable asset for companies in the era of AI, and a trillion-dollar opportunity. This thesis sparked an array of follow-ups, both from Gupta and Garg as well as from others.

Some people, like Gartner’s Afraz Jaffri, believe that using context as an adjective to describe a graph is redundant as a graph implicitly holds context. Others, like Graphwise’s Andreas Blumauer, see context graphs as an evolution that builds upon knowledge graphs, adding time and decision lineage.

Todd Blaschka identifies what he calls the Logic Gap in the context graph narrative: the distance between recording a decision and understanding its meaning. While a knowledge graph defines static relationships, a context graph captures operational reality – decision traces, temporal intelligence, and lineage.

When AI architecture lacks a formal knowledge graph foundation, you encounter three critical failures: identity crisis, hallucinated judgment, and context rot, Blaschka notes. Jessica Talisman elaborates further, arguing that “context graph” is a rebranding, and context graphs are great in theory but will require solid knowledge management foundations to become a reality.

Transform Your AI With A Semantic Layer

Enterprises are pouring millions into AI, but without the right foundation, that investment stalls. Graphwise delivers the knowledge graph and semantic AI infrastructure that make enterprise AI ready to scale, trusted, and built to perform.

Recognized by Gartner, named “Data Integration Innovation of the Year” at the 2025 Data Breakthrough Awards, and listed among KMWorld’s 100 Companies That Matter, Graphwise is the industry’s most comprehensive and validated solution.

Get started with Graphwise today to make generative AI reliable and scalable for your business.

Context Beyond Context Graphs

But there are even deeper issues with the way “context” is used, Talisman expands. When a word becomes a billing unit, the concept associated with the word can quickly lose meaning. Are we discussing context relative to tokens or context designed for AI reliability? Is it a graph? A markdown file? A YAML format or schema tables?

To help disambiguate things, the mission of the W3C Context Graphs Community Group is to develop specifications, vocabularies, and best practices for representing and resolving contextual misalignment between global knowledge representations and local interpretation contexts in decision systems and human–AI workflows.

Athanasius Kircher’s Sephirotic Tree of Life, from Oedipus Aegyptiacus (1652).

Jason Stanley argues that AI agents need five graphs, and nobody has all of them.

Access graphs map who can reach what. Security graphs map what is exploitable and what the blast radius looks like. Context graphs capture decision trajectories so agents can act on precedent. Action graphs model what operations are legal on what objects under what rules. Knowledge graphs represent entities and relationships across the enterprise.

In more practical terms, Andrea Splendiani, Kurt Cagle, the Glean team and Will Lyon share approaches for implementing context graphs. Splendiani and Cagle offer RDF-based alternatives, while Lyon works with Neo4j. The Glean team share their architecture, based on the premise that “you can’t reliably capture the why; you can capture the how“.

The context graph thesis also became the blueprint for the development of the first stable release of Semantica: an open source framework for building context graphs and decision intelligence layers for AI.

Connected Thinking: From civilizational patterns to the next system

A unique journey of exploration, transformation, companionship, and grounding. A series of interactive seminars on foot, reviving the peripatetic school tradition of ancient thinkers in meta-modern times.

● Civilizational Patterns: An Introduction to Macrohistory.

● The Pulsation of the Commons Hypothesis: How commons-based coordination has pulsed through history as an alternative.

● P2P and the Commons: The emerging logic of peer-to-peer as a post-hierarchical coordination model.

● The Next System: What Can We Know? Mapping the contours of what replaces the exhausted form.

Ontologies, Context Graphs, and Semantic Layers: What AI Actually Needs in 2026

Inevitably, the context graph conversation touches upon ontology as well, in the sense of capturing context in a way that both people and AI can reliably use. As Graphlit’s Kirk Marple frames it, entity ontologies are largely solved by existing standards. The real unsolved work is temporal validity, decision traces, and fact resolution.

The current conversation has crystallized around a dichotomy: prescribed ontologies vs learned ontologies. What’s missing from this framing is a third option that’s been hiding in plain sight: Adopt what exists. Extend where needed. Focus learning on what’s genuinely novel.

Even though ontology is considered part of the fundamentals of Information Systems, with 2026 having been proclaimed the year of the ontology, its origins are in philosophy. The entry for Ontology and Information Systems in the Stanford Encyclopedia of Philosophy provides background and references.

In 2026, ontology is trending because AI agents exposed the gap. Years of pipeline-stacking without caring about meaning landed us here: agents failing precisely where semantic understanding was supposed to live. Connectivity without semantics is just faster error, as Frédéric Verhelst notes in In “Own the Ontology or Rent Your Future“.

Verhelst identifies four capability gaps that make agentic AI ungovernable and proposes the Minimum Viable Ontology approach. Following up, he elaborates on the missing contract: why most boards cannot govern what they cannot define, and how to fix this with semantics and ontology.

Ontologies, Context Graphs, and Semantic Layers: What AI Actually Needs in 2026. By Jessica Talisman

The world at large seems to be waking up to the importance of semantics and ontology. Gartner highlighted Data Management, Semantic Layers, and GraphRAG as Top Trends in Data and Analytics for 2026, with Semantic Enrichment recognized as a key capability of Data Management Platforms. Gartner now states that budget for semantic capabilities is non-negotiable.

Bill Inmon, widely recognized as the father of the data warehouse, shared his journey towards semantics and ontology too. Inmon joined forces with Jessica Talisman to introduce some perspectives on ontology, admitting that he never wanted to end up knowing anything about ontology; it was ontology that found him.

Inmon and Talisman followed up with the anatomy of an ontology, where they explore what ontologies look like, how they are structured, and what their defining characteristics and structures are. For people drawn to ontology by the conversation on AI, context graphs and semantic layers, Talisman explores their relationship and what AI needs in 2026.

The shortest path between you and graph insights

Graph visualization and analytics just got a lot easier!

Introducing Linkurious Enterprise Cloud: An online solution that brings powerful graph exploration to anyone, right from a browser.

Create an account, connect your graph database, start the exploration of your data, collaborate with your teammates and share your results, all before lunch.

What else?

• Compatibility with leading graph databases

• Zero IT bottlenecks or infrastructure tasks

• Flexible plans that adapt to your needs

The fastest way to start a graph project today — and the easiest way to scale it tomorrow.

Tooling and Evaluation Frameworks for Ontologies

For non-experts, when it comes to implementing semantic artifacts such as ontologies, semantic work may need its Figma moment. Even when people understand why AI depends on semantics and get the buy-in, Anna Bergevin argues that the tools and process are insufficient for solving this problem.

Bergevin notes that currently, semantics tools are built for experts, not for getting started. She identifies a gap in the market, and believes the parallel success story of how design democratized itself without undermining expertise may be instructive. She is not alone in this observation.

Athanassios Hatzis started a conversation on tooling to visualize ontologies, which soon expanded to include ontology editors. Steve Hedden shared a list of free, open-source RDF & ontology visualization tools. New tools for semantic modeling work such as Termboard, OntoBoom, and OntoView are emerging, while others like gra.fo retiring.

Does Semantic Work Need Its Figma Moment? By Anna Bergevin

Some people may be tempted to get LLMs to write their ontology, but Frédéric Verhelst and Joe Hoeller warn against blindly trusting LLM ontologies – aka “vibe ontologies”. However, like most professionals, knowledge engineers can benefit from using LLMs thoughtfully to assist in their work.

A framework and benchmark for open source LLM-driven ontology construction for enterprise knowledge graphs was presented by Liber AI. More benchmarks for evaluating LLM-generated ontologies were developed, one as a collaboration between LettrIA and EURECOM and another one featuring researchers from German and British universities.

Ontologies are knowledge artifacts, but they’re also software artifacts. Like any software, their quality should be measured in a systematic, operationalizable way. In “Evaluating the Quality of Ontologies“, Neo4j’s Jesús Barrasa and Alexander Erdl reviewed some papers on this topic and implemented some of the ideas they found.

Process Tempo is the missing layer every graph stack needs

Built to accelerate the design, development, and deployment of graph-driven applications, Process Tempo turns your ideas into production-ready solutions faster.

Whether you’re building enterprise knowledge graphs or data intelligence platforms, Process Tempo provides the speed, structure, and flexibility needed to bring your connected ideas to life.

From Retrieval Augmented Generation to Knowledge Augmented Generation

Using ontology in Retrieval Augmented Generation (RAG) is getting traction too. Sergey Vasiliev labels this family of approaches KAG: Knowledge Augmented Generation. Rather than only improving retrieval, the aim is to integrate a knowledge graph as a reasoning substrate. In this view, the graph is not merely a retriever index but a semantic backbone.

In “Enhancing HippoRAG with Graph-Based Semantics“, a team from Graphwise show how an ontology-based knowledge graph boosts the multi-hop Q&A accuracy of a leading schemaless GraphRAG system. Replacing generic graph construction with strict ontologies and structured knowledge graphs transforms HippoRAG from an associative engine into a reasoning engine.

Granter research compared a variety of approaches: standard vector-based RAG, GraphRAG, and retrieval over knowledge graphs built from ontologies derived either from relational databases or textual corpora. Results show that ontology-guided knowledge graphs incorporating chunk information achieve competitive performance with state-of-the-art frameworks.

In Knowledge Augmented Generation, the aim is to integrate a knowledge graph as a reasoning substrate

That’s not to say that other RAG and GraphRAG approaches have gone away. Raphaël MANSUY elaborates on why classic RAG doesn’t work and what to do about it, as a preamble to introducing EdgeQuake: a high performance open source Graph-RAG framework in Rust.

MegaRAG automatically builds knowledge graphs from visual documents. And Graphcore Research published UltRAG: a Universal Simple Scalable Recipe for Knowledge Graph RAG.

A group of Chinese researchers published a survey of Graph Retrieval-Augmented Generation. A systematic survey of GraphRAG, with workflow formalization, downstream tasks, application domains, evaluation methodologies, industrial use cases, and an open source repository.

Google published a guide to building GraphRAG agents with Google’s Agent Development Kit. This hands-on tutorial demonstrates how to create intelligent agents that understand data context through graph relationships and deliver highly accurate query responses.

Steve Hedden explores the rise of context engineering and semantic layers for Agentic AI. He notes that RAG may have been necessary for the first wave of enterprise AI, but it’s evolving into something larger. Neo4j’s Alex Gilmore wrote the Text2Cypher Guide, elaborating on when and how to implement Text2Cypher in agentic applications.

State of the Graph

A comprehensive, up-to-date repository, visualization, and analysis of offerings across the graph technology space.

• Tech professionals exploring graph tools, platforms, and architectures

• Analysts and investors tracking market trends

• Vendors and builders seeking a clear, inclusive map to position their innovations

Learn more

Knowledge Graphs in Software Engineering and Enterprise Architecture

Bala Adithya Malaraju was trying to apply a GraphRAG architecture to his codebase, but running against issues. Then he decided to stop letting LLMs build his knowledge graphs, and adopted the Fixed Entity Architecture. The core idea is simple: instead of letting a LLM discover your ontology from scratch, you define it yourself.

This is just one application of knowledge graphs and ontology in the domain that’s probably seeing the bigest impact from AI: software engineering. There are many more. Amir Hosseini evaluates Codebase-Oriented RAG through Knowledge Graph analysis, using Code-Graph-RAG and gdotv.

GitNexus turns a repository into an AST-driven knowledge graph directly in the browser. session-graph turns scattered AI coding sessions into a queryable knowledge graph. Repository Planning Graph Encoder creates a unified, high fidelity representation for AI-assisted coding.

Repolex offers semantic code intelligence through RDF knowledge graphs.Code review graph creates a local knowledge graph for Claude Code. pr-split decomposes large PRs into a Directed Acyclic Graph of small, reviewable stacked PRs, and gitCGR instantly visualises any GitHub repo as a graph.

Repository Planning Graph Encoder creates a unified, high fidelity representation for AI-assisted coding.

But ultimately software engineering is just one part of Enterprise Architecture. What if ontology could revitalize Enterprise Architecture?

This is the question driving Alberto D. Mendoza’s conversion of ArchiMate 3.2 to an RDF Ontology. Enterprise Architecture frameworks like TOGAF, DoDAF, and FEAF have long used ArchiMate: an open, vendor-neutral, standardized graphical modeling language used to describe, analyze & visualize architectures.

The problem is that after ArchiMate diagrams are created they are flattened, saved as a PDF, and the knowledge it took so long to collect is frozen. But ArchiMate is more than a drawing standard: it’s a formal language with precisely defined element types and relationship semantics.

Elements could be stored in a model that is governed, referenced, and evolves over time rather than recreated from scratch. But EA tools store this information in relational tables, so EA becomes a roadblock. Graphs are the obvious fix. RDF/OWL was designed for rich knowledge representation, so this seems like a natural match.

Connected Data London 2026

10 Years Connecting Data, People and Ideas

Keynote: William Tunstall-Pedoe, the pioneer behind Amazon Alexa

Malcolm Hawker – Thought leader, CDO Profisee

Juan Sequeda – Principal Fundamental Researcher, ServiceNow

Jessica Talisman – Semantic Architect, Founder of The Ontology Pipeline

Book Now

Knowledge Graph Research, Applications and Best Practices

Similar to how software engineering is a premium application domain for knowledge graphs, graph is emerging as the fastest growing segment in AI research. Graph was a significant part of NeurIPS 2025, signifying its growing importance and market share.

Dan McGrath’s findings reinforce this. McGrath tracked the raw growth of graph-related research against the baseline of all AI papers from 2023 to present. The results show a clear acceleration, with a turning point in 2024, when graph became the fastest growing segment in AI research.

Real-world applications abound as well, as shown in Juan Sequeda’s Connected Data London 2025 trip report. A knowledge graph conference where every single talk came from businesses. Not by vendors. Not POCs. Real production deployments with mature architectures and well thought out roles and processes.

Sequeda has been a knowledge graph builder and advocate for decades. He shared 20 lessons from 20 years of building ontologies and knowledge graphs, and he will be back to Connected Data London 2026 as part of an initial lineup also featuring William Tunstall-Pedoe, Malcolm Hawker and Jessica Talisman.

Juan Sequeda’s 20 lessons from 20 years of building ontologies and knowledge graphs

Veronika Heimsbakk wrote a series of posts for data engineers looking to understand knowledge graphs. Kicking off with the motivation – why you should care about knowledge graphs – she elaborates on data engineering ontologies, a few elementary pieces on logic, and shares a translation guide – SPARQL for SQL developers.

Ashleigh Faith also has decades of experience modeling knowledge graphs and ontology. She shares her top 10 modeling tips for ontology and graph. While her tips have a heavy focus on RDF-based graph models, the principles are deep enough to be useful for almost any graph data modeling project.

The debate between the RDF and Labelled Property Graph (LPG) graph data models is ongoing. Sergey Vasiliev explains Property Graph and LPG as structural and applied semantic models, places RDF in its role as a general semantic framework, and formally analyses the relationships between them. He argues RDF is a knowledge representation model and LPG is decision infrastructure.

Niklas Emegård shares a no-ontology hack to show that you don’t need to spend weeks data modeling to start building a RDF knowledge graph, and Pieter Colpaert argues for eventual interoperability– avoiding getting stuck on making trade-off decisions and having to wait for consensus.

Pragmatic AI Training

From Data Literacy to Data Science and Pragmatic AI

For those who want to understand the first principles of AI, and learn how to use it to get results.

• Created through extensive experience

• Designed for busy professionals

• Validated by global leaders

• Delivered on-site

Learn More

Knowledge Graph Tools and Platforms

If you are looking for knowledge graph tools and platforms you can use, there are some resources to help there too.

knwler turns documents into structured knowledge, extracting entities, relationships, and topics. Knowledge Graph Toolkit (KGTK) is a Python library for easy manipulation with knowledge graphs. graflo is a framework for transforming tabular and hierarchical data into property graphs and ingesting them into graph databases.

The State of the Graph is a comprehensive, up-to-date repository, visualization, and analysis of offerings across the graph technology space. The State of the Graph knowledge graph catalog brings together dedicated platforms, infrastructure providers, and knowledge‑centric search and management tools so you can see who is doing what, where they overlap, and where they differ.

TopBraid’s Steve Hedden created Open Knowledge Graphs – a search engine for ontologies, controlled vocabularies, and Semantic Web tools. Ítalo Oliveira created a shortlist of Conceptual Modeling and Linked Data Tools. And Michael Hoogkamer created an interactive taxonomy of semantic modeling concepts.

A Unified Framework for AI-Native Knowledge Graphs. By Fanghua (Joshua) Yu

Taxonomies can be considered as stepping stones for ontologies and knowledge graphs. Heather Hedden shared her insights on what taxonomies are not and taxonomy sources, and Kurt Cagle shows how to use public taxonomies.

Taxonomists have a role in the new world of Generative AI, and Yumiko Saito reflects on it. Kurt Cagle explores how to make taxonomies (and knowledge graphs in general) more friendly for LLMs.

To use LLMs with knowledge graphs, Fanghua (Joshua) Yu proposes Generative Knowledge Modeling (GenKM): a comprehensive methodology introducing a modular four-stage architecture that unifies 40+ existing Graph RAG systems under a common formalism, a generative operator algebra, and the GenKG Lifecycle for end-to-end knowledge graph governance.

The State of the Graph Database Market

The graph database market is growing, with more competition among vendors and more options for users. The State of the Graph catalog of graph databases is an attempt to present the market in a single, structured, vendor‑inclusive view. It aims to enable users to see how graph databases compare across different features.

There are more than 50 graph databases listed on the State of the Graph catalog. But Jason Saltzman, Head of Insights at CB Insights, notes that like cloud infrastructure before it, databases are moving from broad experimentation to standardization around a few critical workloads. As that happens, the market becomes far less forgiving.

Saltzman calls out Neo4j, noting that their momentum reflects scale and defensibility: $200M ARR, 84% Fortune 100 penetration, and accelerating GraphRAG adoption, contributing to one of the highest IPO probabilities CB Insights tracks.

Sudhir Hasbe, Neo4j CPO, elaborates on Neo4j’s evolving architectural evolution in 2025, and shares a roadmap for 2026. Notably, this includes “Ontologies as a First-Class Citizen”: a top-level, independent modeling tool with a repository of use-case-specific samples and native graph schema enforcement. The latest version of Neo4j introduces support for schema as a preview feature.

The State of the Graph database catalog offers a single access point for browsing, comparing, and choosing the offering that is right for your needs.

We saw mobility in the graph database landscape, with new vendors and releases.

TuringDB released Community Version, an open-source edition of its high-performance, versioned graph database. AllegroGraph released v8.5, combining knowledge graphs, vector embeddings, and neurosymbolic reasoning. Memgraph released Atomic GraphRAG Pipelines, implementing sophisticated pipelines as atomic database queries.

SurrealDB released version 3.0, bringing improvements on stability, performance and tooling, developer experience, and building AI agents, while also raising a $23M Series A extension. Vela Partners released a new fork of KuzuDB and added concurrent multi-writer support.

Apparently KuzuDB was acquired by Apple, creating a gap in the graph database ecosystem. In addition to KuzuDB forks, Grafeo is a new embedded graph database built in Rust. Samyama is another distributed graph database written in Rust, which recently released v0.6.1.

Both Grafeo and Samyama highlight LDBC benchmarks results. The Graph Data Council (GDC), formerly known as the Linked Data Benchmark Council (LDBC), is a non-profit organization that defines standard graph benchmarks and fosters a community around graph processing technologies.

There are more benchmarks and updates for the Gremlin ecosystem. LDBC SNB Interactive for TinkerPop is a Gremlin-based implementation of the LDBC Social Network Benchmark (SNB) Interactive v1 workload. TinkerBench is a benchmarking tool designed for graph databases based on Apache TinkerPop. And the Second Edition of Practical Gremlin: An Apache TinkerPop Tutorial was published.

Graph Analytics and Graph AI Updates

The Graph Analytics market is projected to grow from USD 2.41 billion in 2025 to USD 2.92 billion in 2026 (21.61% CAGR), on track to reach USD 9.49 billion by 2032. The State of the Graph catalog for graph analytics offers a single access point for browsing, comparing, and selecting graph analytics tools that match your needs.

A noteworthy entry in the graph analytics market is Google BigQuery Graph. BigQuery Graph, currently in private preview, enables users to query at scale, unify data, and visualize insights, while supporting the Graph Query Language (GQL).

Netflix leverages graph analytics too. The team shared how and why Netflix built a real-time distributed graph and how they created a high-throughput graph abstraction. The next step was the AI evolution of graph search at Netflix, going from structured queries to natural language.

ClickHouse and PuppyGraph introduce the LakeHouse Graph concept: Zero-Copy graph analytics, querying relationships directly on existing data without ETL to a graph database. DuckDB also offers graph analytics now via Onager.

Graphina is a graph data science library for Rust. It provides common data structures and algorithms for analyzing real-world networks, such as social, transportation, and biological networks. The Neo4j blog offers background and examples on some of the most common graph analytics algorithms – Louvain, Jaccard, and PageRank.

“Graph-based Agent Memory: Taxonomy, Techniques, and Applications” presents a comprehensive review of agent memory from the graph-based perspective

Graph AI is being redefined by the advent of graph memory for AI agents. “Graph-based Agent Memory: Taxonomy, Techniques, and Applications” presents a comprehensive review of agent memory from the graph-based perspective. Cognee, an open source AI memory engine that turns data into a living knowledge graph, raised $7.5M seed funding to build memory for AI agents.

Himanshu jha elaborates on a parallel between how Transformers changed sequence modeling and how Graph Transformers might be changing graph learning, framing the shift from GNNs to Graph Transformers through the lens of the Transformer revolution.

Graphbench is a comprehensive graph learning benchmark across domains and prediction regimes. GraphBench standardizes evaluation, and includes a unified hyperparameter tuning framework, and provides strong baselines with state-of-the-art message-passing and graph transformer models and easy plug-and-play code.

Graph Billion- Foundation-Fusion (GraphBFF) is the first end-to-end recipe for building billion-parameter Graph Foundation Models (GFMs) for arbitrary heterogeneous, billion-scale graphs. Central to the recipe is the GraphBFF Transformer, a flexible and scalable architecture designed for practical billion-scale GFMs.

The post Beyond Context Graphs: How Ontology, Semantics, and Knowledge Graphs Define Context. The Year of the Graph Newsletter Vol. 30, Spring 2026 appeared first on The Year of the Graph.

The Ontology issue: From knowledge to graphs and back again. The Year of the Graph Newsletter Vol. 29, Winter 2025 – 2026

George Anadiotis — Tue, 16 Dec 2025 08:06:05 +0000

Enterprise and data architects, data modelers, GenAI adopters, analysts, thought leaders, Graph RAG application builders, Microsoft, Palantir – everyone is talking about ontologies. Why, what is an ontology actually, and how is it related to graphs?

By George Anadiotis

An ontology is an explicit specification of a conceptualization which is, in turn, the objects, concepts, and other entities that are presumed to exist in some area of interest and the relationships that hold among them.

Ontology introduces the semantic foundation that connects people, processes, systems, actions, rules and data into a unified ontology [sic]. By binding real-world data to these ontologies, raw tables and events are elevated into rich business entities and relationships, giving people and AI a higher-level, structured view of the business to think, reason, and act with confidence.

The first paragraph is taken from the entry for Ontology in the Encyclopedia of Database Systems, written by Tom Gruber in 2009. Gruber is credited with the most widely cited definition of ontology in the context of computer and information sciences, introduced circa 1993.

The second paragraph is taken from the blog post introducing Fabric IQ, a new semantic foundation within Microsoft Fabric. It was written by Chafia Aouissi in November 2025. Aouissi is Sr. Director Product Management Fabric IQ at Microsoft.

As usually happens when terms become part of the mainstream discourse, there is a certain tension between rigor and adoption. Case in point – not just Microsoft Fabric IQ, but also Palantir.

Palantir has been floating the term “ontology” as the cornerstone of its architecture for a while. While this has helped popularize the term, as Juan Sequeda notes, at the same time it has also ignited confusion as to what ontology really is.

When Palantir’s CEO, Alex Karp, lashed out at Michael Burry – “Big Short” investor who bet against Palantir and Nvidia – he wasn’t just defending his balance sheet, but also an idea. In Karp’s words, ontology now sits beside chips as the engine of AI. But if we take ontology seriously, Burry’s short might not be wrong, J Bittner notes.

In this issue of the Year of the Graph, we identify ontology and knowledge graph definitions, applications, tools, and educational resources.

The “O” word
From knowledge to graphs and back again
Knowledge graph applications at scale
Ontology and knowledge graph insights, tools and education
Two meanings of “Semantic Layer” and why both matter for AI
Graph databases: growing market, competition, and options
New tools and research

This issue of the Year of the Graph is brought to you by metaphacts, Process Tempo, Linkurious, RDFox, Tentris, Connected Data London, State of the Graph and Pragmatic AI Training.

If you want to be featured in an upcoming issue and support this work, reach out!

Neuro-symbolic AI: The key to truly intelligent systems

Just as you wouldn’t bring half your brain to work, enterprises shouldn’t bring half of artificial intelligence’s capabilities to their architectures. Neuro-symbolic AI combines neural-network technology like LLMs with symbolic technology like knowledge graphs. This integration, also known as ‘knowledge-driven AI’, delivers significant advantages:

● Trustworthy & explainable insights grounded in explicit facts

● Reliable & transparent AI agents

● Grounded LLMs that can assist in complex modeling

If you’re not exploring how knowledge graphs and symbolic AI can augment your

organization’s intelligence—both artificial and actual—now is a good time to start.

Read the full article.

The “O” word

If you talk to people working with data, AI, or enterprise architecture and ask, “what is an ontology?”, you’ll get different answers. For some, ontology is a kind of clever data schema. For others, it’s a business glossary. For others still, the heart of a knowledge graph. They’re all right, and that’s the problem as per Juha-Pekka Joutsenlahti.

In “Demystifying ontologies“, Joutsenlahti gives a brief history of the concept of ontologies in IT and knowledge representation. He explains that different communities adopted “ontology” and bent it slightly towards their own needs, resulting in confusion.

The key to reducing the confusion is to always ask: What is this ontology for? Is it meant to clarify meaning or to define data structure (or both)? Once we make that distinction explicit, much of the mystery starts to disappear.

“The O word: do you really need an ontology?” was published in 2019. Before GenAI was a thing, Mark Hall made a compelling case for ontologies and offered an explanation as to why isn’t everyone doing this.

Today, as Ole Olesen-Bagneux notes, ontologies are once again hot because they are key to succeeding with AI: ontologies provide context for AI to perform better. Thus, we are seeing the re-introduction of knowledge engineering as if it were new.

How Data Models and Ontologies Connect to Build Semantic Foundations

Knowledge Management and the Library Sciences, from which taxonomies, ontologies, and knowledge graphs were born, are well-established disciplines, as Juha Korpela notes in “How Data Models and Ontologies Connect to Build Your Semantic Foundations“.

Korpela points out that people who have traditionally worked with ontologies and knowledge graphs have not been communicating much with domains such as data modeling, but the exchange would be meaningful.

Data modelers focus on the technical implementation of data solutions, thus following a path from Conceptual to Logical to Physical modeling. Even if concept models and ontologies are different, as Jessica Talisman notes, there is overlap. Conceptual modeling may be used to build an ontological foundation that acts as the context provider for agents and chatbots as well as humans.

Bring your graph ideas to life!

Building a knowledge graph? You’re going to want people to use it. We can help!

Process Tempo Jupiter makes it easy to build beautiful, production-ready applications on top of your graph without writing code. Now you can bring your graph ideas to life using skills you already have.

Jupiter supports multi-modal, read and write, between graph databases, relational databases and Rest-based APIs. No ETL required!

Neo4j, Memgraph, PuppyGraph, OpenCypher

Databricks, Snowflake, Big Query, MySQL, Postgres, Dremio

Ready to get started? Get on our calendar to learn more: calendly.com/processtempo

From knowledge to graphs and back again

Another term that is overloaded, being used in different contexts to mean different things, is “knowledge graph”. For an elaboration of the definition of a knowledge graph, the connection to ontology, and how they become the essential truth layer for Pragmatic AI, check this conversation between Tony Seale and George Anadiotis.

Taewoon Kim has a go at defining knowledge graphs and offering a practical guide of the different options for implementing them. Nicolas Figay summarizes the insights from different viewpoints on what a knowledge graph really is.

Like ontologies, what has largely contributed both to popularizing and creating confusion around knowledge graphs is their use for GenAI, specifically to support LLMs in Graph RAG. As Fanghua (Joshua) Yu notes, even when talking about knowledge graphs in Graph RAG use cases, there are different KG types to consider.

Knowledge Graphs and GenAI: When the Complexity Is Worth It

But is using knowledge graphs with GenAI always a good idea? The complexity may not always be worth it, Dave Bechberger argues. You may not always need Graph RAG. While graph-based enhancements improve information organization & reasoning, they also come with a heavy token cost.

Graph RAG is a data engineering challenge, not just an LLM trick, Fanghua (Joshua) Yu notes. Plus, Graph RAG does not always outperform “vanilla” RAG. If you are going to be using Graph RAG, however, there are some useful frameworks, applications and analyses to consider.

Flexible GraphRAG and ApeRAG are configurable open source frameworks for Graph RAG. Niklas Emegård built an open source semantic Graph RAG application that can be used as a template for projects. Sergey Vasiliev explores how graph data science and analytics can help power Graph RAG applications. And Benito Martin shares how he built a Biomedical GraphRAG system, combining knowledge graphs with vector search.

The shortest path between you and graph insights

Graph visualization and analytics just got a lot easier!

Introducing Linkurious Enterprise Cloud: An online solution that brings powerful graph exploration to anyone, right from a browser.

Create an account, connect your graph database, start the exploration of your data, collaborate with your teammates and share your results, all before lunch.

What else?

• Compatibility with leading graph databases

• Zero IT bottlenecks or infrastructure tasks

• Flexible plans that adapt to your needs

The fastest way to start a graph project today — and the easiest way to scale it tomorrow.

Knowledge graph applications at scale

Today, knowledge graph applications in production are scaling up. Walmart uses its People.AI knowledge graph to power recommendation tools. This knowledge graph includes 1.6 million nodes and 83 million edges, representing entities such as job titles, openings, associates, applicants, and skills.

Barclays is scaling knowledge graphs and knowledge-based AI in the financial crime and KYC domain. They are using semantic reasoning to automatically enrich knowledge graphs with new facts based on expert knowledge to deliver improved operational efficiency, reduce compliance risk and improve customer experience.

GitLab is building what they call the GitLab Knowledge Graph. It’s a framework written in Rust to turn your codebase into a live, embeddable graph database that can be used for code retrieval and navigation, impact analysis and architecture visualization.

RTVE-Grafo: The knowledge graph of the Spanish audiovisual archive

RTVE, the Spanish national radio and television public broadcaster, has created a knowledge graph that structures information by understanding the relationships among various audiovisual content as well as any potentially linked objects. It unifies data, making it accessible and comprehensible for both machines and people.

Wiz, the cybersecurity phenomenon recently acquired by Google for $32 billion, has redefined cloud security by providing comprehensive visibility across multi-cloud environments. Their success is based on a massive, constantly evolving security graph that spans customers’ entire cloud footprints.

Microsoft introduced the Sentinel graph to bring relationship-aware context to Microsoft Security across Defender and Purview. This way defenders and AI can see connections, understand the impact of a potential compromise, and act faster across pre-breach and post-breach scenarios.

Graph-Wide scanning is a technique for solving advanced cyber threats. In one example, a query on a 150 billion-edge graph scanned a mind-boggling 123 trillion edges. It took time, but it found fewer than 4,000 answers, highlighting the power of finding the critical few from the overwhelming many.

Ontology and knowledge graph insights, tools and education

A key finding of the State of the Graph survey for 2025 is that knowledge graphs and graph databases are driving adoption, but guidance and training are still critical. Here are some pointers to knowledge graph and ontology educational resources.

Connected Data London 2025 was a gathering of top minds and practitioners in this space, featuring use cases, innovation and educational content from the likes of Airbus, AstraZeneca, AWS, Barclays, Bloomberg, Netflix, Nvidia, SAP, ServiceNow, S&P, Vodafone and more. You can follow CDL for trip reports and teasers, or dive into the content with a Remote Access Ticket and a special 15% discount.

metaphacts recently published their Semantic Modeling Guidelines for Knowledge Engineers. These semantic modeling guidelines are designed for beginners as well as advanced modelers, offering a step-by-step introduction to semantic modeling concepts, key elements and practical techniques.

Kurt Cagle shares tips for building knowledge graphs, noting that the hard part of building a knowledge graph is not the technical aspects, but identifying the types of things that are connected, acquiring good sources for them, and figuring out how they relate to one another.

It is better to create your own knowledge graph ontology, possibly building on existing upper ontologies, than it is to try to shoehorn your knowledge graph into an ontology that wasn’t designed with your needs in mind.

Becoming an Ontologist

In “Becoming an Ontologist“, Cagle notes there is a surge in interest in the profession of ontologist. Some of it can be attributed to the fact that people are beginning to realize the value of knowledge graphs, but there is also the opportunistic element here. Like many other fields in the past, ontology work is seen as a ticket to big money. But perceptions and reality are not necessarily aligned.

Dean Allemang shares his insights on a day in the life of a working ontologist. Building ontologies is actually the last thing on the list, as there isn’t much spent on that compared to other tasks. Allemang notes that “ontologist” is going to be a much more sought after skill in the near future.

Check also these tools for visualizing, editing and creating ontologies and conceptual modeling and Linked Data. Robert Sanderson shared his 10 design principles for knowledge graphs and ontology, and Giancarlo Guizzardi shared a tutorial on the Unified Foundational Ontology. And the AIOTI published a report on the different Data to Ontology mapping tools available.

For more in-depth education:

Pragmatic AI Training: A holistic AI education program, including modules on knowledge graphs, Graph RAG and ontology design
The Knowledge Graph Academy: learn how to build and scale knowledge graphs through a unique program led by global experts

Do you need a graph technology that scales without sacrificing performance?

RDFox® is one of the most advanced RDF triple stores on the market. Built on world-leading research from the University of Oxford, RDFox® stands out through:

● Highly performant, scalable in-memory architecture

● Advanced rule-based reasoning for dynamic graph enrichment

● An optimised memory footprint for on device, server, or cloud

RDFox® is trusted by a growing list of the world’s largest brands across financial services, manufacturing, automotive, retail, and big tech.

Powering what is possibly the world’s largest graph deployment across tens of millions of Samsung smartphones – see what all the fuss is about here.

Two meanings of “Semantic Layer” and why both matter for AI

Another concept everyone is talking about, but not necessarily meaning the same thing, is semantic layers. On the analytics side, semantic layer usually refers to something we’ve seen before. It’s the business-friendly model that sits between raw data and BI tools. A governed way to define joins, metrics, and consistent logic.

This version of semantics is about trust. It makes sure when someone (or something) asks, “What’s our total revenue last month?” the answer is accurate, governed, and consistent.

Case in point: the Open Semantic Interchange (OSI). Snowflake, Salesforce, dbt Labs and other vendors announced they are working on what they claim will be a universal standard for how business data is defined and shared across platforms.

Semantic Layers and GraphRAG are essential for Trustworthy AI

But there’s another camp using the same phrase and they’ve been doing it far longer. For ontologists, RDF, OWL and JSON-LD are open standards for exchanging semantic data. To them, a semantic layer isn’t a metrics model, it’s a knowledge model. It’s about representing meaning, relationships, and context across systems.

This version of semantics is about understanding. It connects definitions and relationships, providing the context AI uses to make sense of information.

Case in point: Semantic Layers and GraphRAG are essential for Trustworthy AI, Andreas Blumauer points out. Agents need a semantic layer, Anthony Alcaraz chimes in.

Query more, wait less. Try Tentris today

Tentris is a new disk-based RDF graph database that delivers blazing fast querying performance with highly efficient memory usage, turning analytics that once took days into results within minutes or seconds. Built for real time workloads and complex data, it solves tasks others cannot.

Start exploring Tentris today!

Subscribe to the Year of the Graph Newsletter

Keeping track of all things Graph Year over Year

Graph databases: growing market, competition & options

Moving to graph database updates, let’s open with another piece of news from Microsoft. In Microsoft’s Q1 2026 earnings call, it was reported that Cosmos DB business grew over 50% YoY. Cosmos DB is a multi-model database, and we don’t know the extent to which graph contributed to its growth.

However, this signal is one of many pointing towards growth for graph databases. According to Fortune Business Insights, the global graph database market size is projected to grow from $2.85 billion in 2025 to $15.32 billion by 2032, exhibiting a CAGR of 27.1%.

Microsoft, AWS, Google, and Oracle were named as leaders in the 2025 Gartner® Magic Quadrant for Cloud Database Management Systems. All superscalers have a graph database offering with Cosmos DB, Amazon Neptune, Google Spanner Graph, and Oracle Graph, respectively.

Gartner Magic Quadrant for Cloud Database Systems 2025

Neo4j was the only pure-play graph database to be listed in the same Magic Quadrant as a niche player. Neo4j also announced Fleet Manager, a single control plane for all Neo4j deployments, as well as the release of Neo4j Graph Analytics for Snowflake in the Snowflake Marketplace.

But there has also been churn in the graph database market. Dgraph was acquired by Istari Digital to strengthen data foundation for AI and engineering. And the KuzuDB embedded open source graph database has been abandoned by its creator and sponsor Kùzu Inc.

LadybugDB and RyuGraph are new forks of the KuzuDB codebase aiming to pick up where KuzuDB left off. Furthermore, GraphLite emerged as a new open source graph database for embedded processes, and FalkorDB introduced FalkorDBLite – both aiming to fill in the embedded graph database void left by KuzuDB’s departure.

Challengers such as QLever, RushDB, TuringDB, TypeDB, and DuckDB with its DuckPGQ extension are emerging. It seems like the graph database market pie is growing, and the competition for a piece of it is intensifying.

Finally, another signal pointing in the direction of growth for the graph market: Linkurious, who just published the 2026 update of their Graph technology landscape, are getting acquired by Nuix in a €20M deal.

New tools and research

Wrapping up with a roundup of new tools and research. Focusing on graph research and innovation in NeurIPS, one of the leading AI conferences, shows that graph is a significant part of NeurIPS 2025, supporting its growing importance and market share.

GraphFrames represents the natural evolution of GraphX, the native library for cluster-based graph processing on Apache Spark – one of the most powerful engines for large-scale data processing in distributed computing. GraphFrames has been revived after being dormant for a while, with version 0.10.0 recently released.

Brahmand is an open source Graph Engine built on top of ClickHouse. It extends ClickHouse with graph modeling and OpenCypher, merging OLAP speed with graph analysis.

Gephi Lite v1.0 was recently released. Gephi Lite is a lighter, web-based version of Gephi, used for visual network analysis. Cosmograph, a single-node web-based tool used to visualize graphs, released v.2.0.

Ontology-Guided open-domain knowledge extraction with LLMs

ODKE+ supports Ontology-Guided open-domain knowledge extraction with LLMs, and GraphQA introduces an open source agent for asking graphs questions. Tree-KG is an expandable knowledge graph construction framework for knowledge-intensive domains.

Text2KGBench-LettrIA is a refined benchmark for Text2Graph systems, attempting to address the question of whether LLMs are good at populating knowledge graphs based on a set of text documents. RAGE-KG explores the state of the art and beyond in integrating Retrieval-Augmented Generation with Knowledge Graphs as well as the synergies between Large Language Models and the Linked Open Data ecosystem.

KGGen extracts knowledge graphs from plain text with language models. Exploring Network-Knowledge graph duality is a case study in agentic supply chain risk analysis. And GraphPFN is an attempt to create general-purpose graph foundation models from tabular foundation models.

Last but not least: the Connected Data Knowledge Graph v0.1 was released – an open knowledge graph for the community by the community.

The post The Ontology issue: From knowledge to graphs and back again. The Year of the Graph Newsletter Vol. 29, Winter 2025 – 2026 appeared first on The Year of the Graph.

Graph is the new star schema. The Year of the Graph Newsletter Vol. 28, Autumn 2025

George Anadiotis — Mon, 22 Sep 2025 16:33:50 +0000

Are graphs really the new star schema? What do graphs look like to non-insiders, and what is it that attracts them to the graph community, methodologies, applications, and innovation?

By George Anadiotis

In the process of connecting with Connected Data London 2025 speakers, we found ourselves engaging in conversations reflecting and elaborating on the origins and evolution of the community and the technology it converges on: Knowledge Graphs, Graph Analytics, AI, Data Science, Databases and Semantic Technology.

The way we described it was – a core of early adopters that have been around for what can be more than a decade by now, plus a growing segment of newcomers who stumbled upon graphs prompted either by GenAI or by the realization that graph is the best way to model connectedness.

Everything is connected. Graphs exist to model connections. This is why graphs are everywhere. Today, there are more data points to be connected than ever before. There are also more connections made faster than ever before. And people are realizing and leveraging the importance of these connections more than ever before.

This is why there is an explosion in adoption, thought leadership, tools, features, applications and methodologies around graphs. This issue of the Year of the Graph reflects this. Even though it’s only been 3 months since the previous one, and the targeted curation approach hasn’t changed, it’s the longest issue to date.

This issue caters to everyone – from newcomers to experts, and from strategic thinkers and modelers to engineers and scientists. From graph support on Microsoft Azure, GitLab, Netflix, S&P and SAP, to semantics and agents on Databricks and Snowflake, ontologies, knowledge graphs and AI, graph transformers, and science breakthroughs.

To meet the leaders and innovators shaping Knowledge Graph, Graph Databases, Semantic Technology and Graph Analytics / Data Science / AI, come to Connected Data London 2025 on November 20 – 21.

- Graph is the new star schema
- Graphs power Systems of Intelligence
- A unified semantic knowledge graph for Enterprise AGI
- Defining and building ontologies
- Getting started with knowledge graphs
- Adopting, building and populating knowledge graphs
- Knowledge graphs and AI: a two-way street
- The state of GraphRAG
- Multimodal graphs
- Graph databases grow and evolve
- LPG vs. RDF, OWL vs. SHACL
- Graph AI: GNNs, graph transformers and foundational models
- Graph science: Strong perfect graphs, the new Dijkstra’s algorithm and convergent neural networks

This issue of the Year of the Graph is brought to you by G.V(), metaphacts, Linkurious, and cognee.

If you want to be featured in an upcoming issue and support this work, reach out!

Announcing the State of the Graph project

A comprehensive and up to date repository, visualization, and analysis of all offerings in the graph technology space.

Graph is the new star schema

Is graph really the new star schema? Irina Malkova, VP Product Data & AI at Salesforce, thinks so. Before AI, Malkova didn’t feel like graph metadata was an ROI positive investment for her team – even though they never looked too closely.

Now, however, Malkova realizes that agents can’t be autonomous unless data is structured as a graph. Just a few days after Malkova voiced this realization, her post attracted graph leaders who shared more background and context.

As she puts it, we’re living through the third wave of graphs, now driven by the need to feed data to AI agents. Malkova wonders what it would be like if graph insiders and newcomers all reimagined their jobs by learning from each other.

What do graphs and the data warehouse star schema have in common? They are ubiquitous, and they help unlock value for organizations. Unlike the star schema, however, graph data models are flexible and can unambiguously model semantics. Put another way – graph schema is the star of schemas.

As Charles Betz, VP, Principal Analyst at Forrester notes, without well-defined processes and resilient architectures, the acceleration that GenAI brings risks amplifying chaos instead of streamlining workflows.

The key lies in building a robust IT operating model that integrates GenAI into the fabric of management systems. Graph databases and retrieval-augmented generation are foundational technologies for this transformation.

Graphs represent entities and relationships flexibly, allowing GenAI to reason across complex data landscapes. By investing in graph-based knowledge infrastructure, organizations can unlock the full potential of AI while ensuring transparency, traceability, and alignment.

Matan-Paul Shetrit, Director of Product Management at Writer, takes this one step further, envisioning graphs as the orchestration layer for firms of the future. In the traditional enterprise, coordination was manual. In the hybrid enterprise, coordination becomes programmable.

This reframes the org chart. It’s no longer the primary map of how work flows. The orchestration graph is: the dynamic, often invisible network of people, agents, and systems connected by delegation logic, execution loops, and escalation paths.

No matter what graph technology you work with, G.V() makes you more productive

G.V() is a graph database client and IDE that empowers you through every task:

Write, execute, and profile queries
Inspect your data model
Explore your data with high-performance graph visualization
Add or edit data on the fly

As the most widely compatible graph database IDE, G.V() supports 20+ tools, including Amazon Neptune, Google Spanner Graph, Neo4j, and JanusGraph – and now with GQL support for Ultipa Graph.

Try it out for yourself and start querying your database in less than 5 minutes: gdotv.com

Graphs power Systems of Intelligence

Graphs are evolving alongside GenAI, influencing as well as being influenced by it. According to the Gartner 2025 AI Hype Cycle, Generative AI capabilities are advancing at a rapid pace and the tools that will become available over the next 2-5 years will be transformative.

Al investment remains strong, but focus is shifting from GenAl hype to foundational innovations like Al-ready data, Al agents, Al engineering and ModelOps. The rapid evolution of these technologies and techniques continues unabated, as does the corresponding hype, making this tumultuous landscape difficult to navigate.

These conditions mean GenAI continues to be a top priority for the C-suite. Knowledge Graphs are a key part of this, positioned on the slope of enlightenment. Knowledge graphs are also the foundation for Systems of Intelligence.

Systems of Intelligence is a term coined by Geoffrey Moore and referenced by SiliconANGLE & theCUBE analysts David Vellante and George Gilbert in their analysis on how Snowflake navigates new competitive dynamics.

Systems of Intelligence are the linchpin of modern enterprise architecture because AI agents are only as smart as the state of the business represented in the knowledge graph. If a platform controls that graph, it becomes the default policymaker for “why is this happening, what comes next, and what should we do?”

A unified semantic knowledge graph for Enterprise AGI

Vellante and Gilbert also zoom in on the importance of semantics in their analysis of how Ali Ghodsi’s data intelligence playbook for Databricks is turning data into agentic advantage. Understanding the strategy lies in recognizing how Systems of Intelligence, Systems of Engagement, and Systems of Agency create a flywheel:

User Intent Feeds Semantics: Each question ultimately enriches the catalog with richer context, adding meaning to the data so that others can extract more value from it.
Semantics Feed Agents: Richer context arms agents to make better decisions and ultimately to act autonomously.
Agents Create Outcomes: Agents deliver outcomes more effectively aligned with business objectives.

Ali Ghodsi has made the stakes clear, calling owning semantics “existential”. This is the cornerstone of the company’s Data Intelligence strategy. Without harmonized, governed & enriched semantics, dashboards, agents, or operational applications can’t integrate and function.

The goal is to enable agents that not only know what happened, but can also infer why, predict what’s next & prescribe what to do. This is the holy grail of enterprise AI – the foundation for “Enterprise AGI.” Getting there means abstracting beyond RDBMSs and tables, to a unified semantic knowledge graph.

Strategic thinkers realize the importance of owning semantics, and semantic layers matter more than ever in the AI era. This is part of the reason why people are talking about semantic layers – but what is a semantic layer, actually?

Connected Data community members Sofus Macskássy, Jessica Talisman, Juan Sequeda and Andreas Blumauer approach semantic layers from a multitude of angles, sharing definitions and guidelines.

Why AI alone can’t solve all your data problems

LLMs often “hallucinate” because they rely on broad, statistical patterns, not your business’s unique knowledge. Grounding AI in your semantic models ensures outputs are accurate and trustworthy by leveraging logical connections within enterprise data.

metis is a knowledge-driven AI platform that transforms disconnected data into business value. With metis, you can create and manage semantic models with AI, design and deploy custom conversational agents, and combine tools for summarization, query execution and more.

Use reliable, trustworthy AI that truly understands your business.

Discover metis today!

Defining and building ontologies

Defining semantic layers could be called meta-semantics, which is a bit ironic, but at the same time necessary. Precise definitions is what semantics is all about, and with renewed interest and newcomers in the field, confusion, hype and heated arguments are inevitable side effects.

The same can be said about ontology. When CTOs ponder “what is an ontology” and experts weigh in, the conversation can get pretty technical pretty fast. Others focus on more practical aspects, such as what organizations can learn from the poster child for ontology in use: Palantir.

This points towards another perpetually ongoing conversation: even assuming people agree on what an ontology is, what is the best way to build one? Is ontology engineering really a mess? And why are ontology and data architecture teams solving the same problems with different languages?

Palantir has its own, proprietary definition and implementation of ontology. Some people believe that this is counter-productive. Standards such as OWL, RDF, SKOS and SHACL exist, so there is no reason to reinvent the wheel. Others think that while interoperability is good, those standards are not always fit for purpose, and group think limits evolution.

There is growing consensus that ontologies matter, hard as they may be to develop. Joe Hoeller argues ontology development is hard because it relies on subject matter experts who bring deep understanding of operational language, workflows, and terminology but typically lack the formal training needed to represent that knowledge within rigorous frameworks.

Top Level Ontologies (TLOs) such as the Basic Formal Ontology (BFO), Common Core Ontology (CCO), or SUMO (Suggested Upper Merged Ontology) are such rigorous frameworks. J Bittner explains that TLOs help with ROI in two ways.

First, avoiding reinvention – you don’t need to debate or rebuild basic categories for every project. Second, when domains need to interoperate – finance with logistics, compliance with operations, healthcare with insurance – the shared foundation dramatically lowers the cost of integration.

Using LLMs to support ontology development is something that people like Tony Seale have been advocating. Stardog CEO & Founder Kendall Clark claims that building ontologies leveraging foundational models is “magic” that works via prompt scaffolding, symbolic alignment, formal encoding and iterative validation.

However, Semantic Arts CEO & Founder Dave McComb challenges the notion that getting an LLM to create even a lousy ontology is a good starting point. This, he claims, is premised on the assumption that we need to build a lot of ontologies, as every project needs an ontology.

What we need, McComb argues, are orders of magnitude fewer ontologies. You really see the superpowers of ontologies when you have the simplest possible expression of complex concepts in an enterprise.

This belief is reflected in gist, an open-source, business-focused ontology actively developed by Semantic Arts. Its lightweight design and use of everyday terminology has made it a useful tool for kickstarting domain ontology development in a range of areas. gist is now aligned with BFO.

Regardless of how you choose to build your ontology, tools and methodologies to help do exist. From books such as Introduction to Ontology Engineering, to tools such as the open source Spreadsheet-based Ontology Maker and OntoAligner, a comprehensive modular and robust Python Toolkit for ontology alignment.

Sneak peek: Graph visualization and analytics, reimagined for the cloud

Founded in 2013, Linkurious helps Global 2000 companies and government agencies turn complex connected data into clear insights.

Now, we’re bringing our flagship graph visualization and analytics solution to the cloud. Linkurious Enterprise Cloud (just weeks away from launch…be among the first to try it!) is the most user-friendly, flexible, and scalable way to explore your graph data. In minutes, create an account, connect your graph database, and explore relationships at scale – no infrastructure or maintenance required.

Get early access – join the waitlist for a free trial.

Getting started with knowledge graphs

A semantic layer built with knowledge graphs increases the inherent value of data by enabling faster data integration and improving data quality and trust with contextual relationships and shared models. This is why organizations need to adopt the new paradigm of ECL (extract, contextualize, load) instead of traditional ETL (extract, transform, load) for improved ROI on AI, argues Sumit Pal.

Businesses must ground their AI in knowledge graphs, Joe Hoeller chimes in. First, because your tabular data is dumb, even though your business isn’t. And second, because as opposed to predictive (statistical) inference based on LLMs, ontological (logical) inference based on knowledge graphs is deterministic and explainable.

You may think that knowledge graphs are too complex to implement, or that they need massive datasets. These are just some of the common misconceptions around knowledge graphs that Vasilije Markovic tries to address. But even if you are new to graphs, there are lots of resources to get you started.

In “Relational to Graph“, Frank Blau offers a guide to Graph Thinking. In the “Graph Tech Demystified” series, Paco Nathan shares how to get up to speed on graph fundamentals. Max De Marzi shares graph modeling mastery tips. And if you are looking to land a Knowledge Graph Engineer role, Thomas Thelen compiled a list of interview questions.

Building a knowledge graph sounds great, but jumping in too early without answering some key foundational questions can lead to wasted time and over-engineering. Sabika Tasneem shares 15 questions to help you start smart, whether you’re building a simple internal graph or planning a complex GenAI-powered system.

Implementing an enterprise knowledge graph is a multi-phase journey. As projects move from an initial proof-of-concept to a fully productionalized, multi-domain graph , costs accumulate. Joe Hoeller breaks down the typical stages (PoC, pilot, and full enterprise deployment) and the costs entailed in each.

Cognee turns any data into a queryable knowledge graph backed by embeddings

Cognee turns unstructured, structured and semi-structured data into a queryable knowledge graph backed by embeddings.

Cognee retrievers blend vector similarity with graph traversal for precise, multi-hop answers and reproducible context – so agents reason and remember with structure. Add Cognee’s enrichment layer, time-awareness, auto-optimization, and its new UI for an even better experience. For teams building domain-aware agents, copilots, and search for knowledge-heavy domains.

Cognee is open source, with a hosted version – cogwit. Try it.

Adopting, building and populating knowledge graphs

Knowledge graph adoption is peaking. There are more people wanting to build knowledge graphs, and more tools and approaches to do this than ever before.

On SAP, you can now use semantic querying with the SAP HANA Cloud knowledge graph. GitLab’s knowledge graph can be used for codebase RAG, code navigation, impact analysis and architecture visualization. Graph-Code is an open source graph-based RAG system for any codebase.

Synalinks is a Keras-based neuro-symbolic framework that bridges the gap between neural networks and symbolic reasoning. SynaLinks latest release 0.3features optimized and constrained knowledge graph extraction and retrieval, integration with agents, Neo4j support, Cypher query generation and automatic entity alignment.

Cognee is a modular framework for end-to-end knowledge graph construction and retrieval. A joint post by the cognee and Kuzu teams shows how to transform relational data into a knowledge graph. Similarly, Amber Lennox shares how to go from raw data to a knowledge graph with SynaLinks.

There’s no shortage of tutorials on other tools either. Gal Shubeli shows how to build a knowledge graph from structured & unstructured data using FalkorDB and Graphiti. Thu Hien Vu shares how to extract knowledge graphs from text with GPT4o, and Alain Airom builds a knowledge graph from documents using Docling.

OntoCast is an open source framework for creating knowledge graphs (extracting semantic triples) from documents using an agentic, ontology-driven approach. It combines ontology management, natural language processing, and knowledge graph serialization to turn unstructured text into structured, queryable data.

AutoSchemaKG is a framework for fully autonomous knowledge graph construction that eliminates the need for predefined schemas. It leverages LLMs to extract knowledge triples and induce comprehensive schemas directly from text. AutoSchemaKG is cutting edge research, with the code released on GitHub.

Blue Morpho markets a system that turns PDFs and text files into knowledge graphs. iText2KG, an open source Python package designed to incrementally construct consistent knowledge graphs with resolved entities and relations, can now build dynamic knowledge graphs.

Russel Jurney introduces the emerging field of semantic entity resolution for knowledge graphs, which uses LLMs to automate the most painful part of building knowledge graphs from text: deduplicating records.

Andrea Volpini shares a notebook exploring semantic entity resolution & extraction using DSPy and Google’s new LangExtract library, and Prashanth Rao offers a gentle introduction to DSPy for graph data enrichment.

Subscribe to the Year of the Graph Newsletter

Keeping track of all things Graph Year over Year

Knowledge graphs and AI: a two-way street

In the real world, using predefined entities and relationships while cleaning up and resolving duplicates and flagging inconsistent sources is a requirement for building knowledge graphs.

As Panos Alexopoulos notes, these are the types of knowledge graph quality issues that hamper downstream applications. And trying to automate knowledge graphs may also end up having unforeseen consequences.

The authors of “Knowledge Graphs and LLMs in Action” believe that knowledge graphs and LLMs can work together. They show how to model knowledge graphs based on business needs and unstructured text data sources, how to leverage ontologies, taxonomies, structured data, machine learning algorithms and reasoning.

The relationship goes the other way round, too. As part of their interpretability research, Anthropic introduced a new method to trace the “thoughts” of a large language model. The approach is to generate attribution graphs, which (partially) reveal the steps a model took internally to decide on a particular output.

Anthropic open sourced a library that supports the generation of attribution graphs on popular open-weights models, releasing a frontend to explore graphs. Michael Hunger wrote a script to import the graph json into Neo4j, and Srijan Shukla open sourced code to transform Claude’s hidden memory into interactive knowledge graphs.

In the world of LLMs, the term “context engineering” has been getting traction. LangChain’s CEO Harrison Chase defines context engineering as “building dynamic systems to provide the right information and tools in the right format such that the LLM can plausibly accomplish the task”.

As Jérémy Ravenel notes, context without structure is narrative, not knowledge. And if AI is going to scale beyond demos and copilots into systems that reason, track memory, and interoperate across domains, then context alone isn’t enough. We need ontology engineering.

Context engineering is about curating inputs: prompts, memory, user instructions, embeddings. It’s the art of framing. Ontology engineering is about modeling the world: defining entities, relations, axioms, and constraints that make reasoning possible. Context guides attention. Ontology shapes understanding.

Knowledge graphs backed by ontologies are the ultimate context layer for LLMs. Knowledge graphs excel at providing structured, semantic context to LLMs by organizing information as interconnected entities and relationships, making them great options for Memory and Retrieval, as Anthony Alcaraz notes.

Agentic knowledge graph construction and temporal graphs

Agentic knowledge graph construction is the latest in automated knowledge graphs. The idea Anthony Alcaraz promotes based on a tutorial by Andrew Ng and Andreas Kolleger is to deploy an AI agent workforce, treat AI as a designer, not just a doer, and use a 3-part graph architecture to augment humans instead of replacing them.

The authors of the book “Building AI Agents with LLMs, RAG, and Knowledge Graphs” aim to equip data scientists to build intelligent AI agents that reason, retrieve, and interact dynamically, empowering them to deploy AI solutions. They dedicate a chapter to creating and connecting a knowledge graph to an AI Agent.

Google Cloud released Google Agentspace, which provides a single platform to build, manage, and adopt AI agents at scale for individuals, teams, and enterprises. Google Agentspace is powered by a knowledge graph, built on Spanner Graph.

FinReflectKG is a concrete example of agentic construction and evaluation of financial knowledge graphs, claiming to be the largest open source financial knowledge graph built from unstructured data.

OpenAI released a hands-on guide for building Temporal Agents with knowledge graphs and performing multi-hop retrieval directly over those graphs. While the cookbook focuses on OpenAI models and some other specific tooling, the underlying framework and logic are model-agnostic and easily adaptable to other stacks.

Another approach is shared by Fareed Khan in “Building a Temporal AI Agent to Optimize Evolving Knowledge Bases in Modern RAG Systems“. Khan shows how to create an end-to-end temporal agentic pipeline that transforms raw data into a dynamic knowledge base, and build a multi-agent system to measure performance.

Temporal graph modeling is what TGM focuses on. TGM is a research open source library designed to accelerate training workloads over dynamic graphs and facilitate prototyping of temporal graph learning methods. It natively supports both discrete and continuous-time graphs.

The state of GraphRAG

And what about GraphRAG? Just over a year ago, GraphRAG was the hottest topic in AI. GraphRAG is an emerging set of techniques to enhance retrieval-augmented generation by integrating knowledge graphs, using their structured nature to provide richer, more nuanced context than standard vector search could offer.

Several architectural blueprints for harnessing these graphs to capture the complex relationships between entities were laid out, with the goal of producing more accurate and contextually aware AI-generated responses. Since then, Ben Lorica has been watching for signs of these techniques taking root in practice.

While evidence of widespread adoption is scarce, forward-looking applications are emerging. In agentic AI systems, the graph is evolving from a simple data source for retrieval into a foundational map for reasoning and coordination. The true value of the graph-centric approach becomes clear when applied to agentic AI, Lorica argues.

GraphRAG is still hot. Neo4j published the Developer’s Guide to Graph RAG and Essential GraphRAG book. Avi Chawla shares a visual explainer of RAG vs Graph RAG, while a empirical study analyses when and how to use knowledge graphs for RAG. Anthropic’s Contextual Retrieval claims to decrease RAG retrieval error rate by 67%.

There are more new GraphRAG variants too.

HippoRAG takes cues from the brain to improve LLM retrieval.
Graph-R1 combines GraphRAG with Reinforcement Learning.
DRAG introduces a novel distillation framework that transfers RAG capabilities from LLMs to SLMs through evidence-based distillation and Graph-based structuring.
HiRAG uses hierarchical clustering to link disparate topic clusters to enhance global reasoning.

Andreas Kolleger highlights innovative approaches from the GraphRAG Track at AI Engineer World’s Fair 2025, and Ben Lorica shares 5 breakthroughs you should know about in RAG Reimagined.

Sergey Vasiliev argues GraphRAG doesn’t lack ideas, but it struggles to scale up. To counter this, Huawei researchers published a pragmatic case study in balancing scalability with reasoning depth in GraphRAG systems.

Multimodal graphs

A topic that’s gaining momentum in GraphRAG and beyond is multi-modality. RAG-Anything is an all-in-one RAG system that leverages multimodal knowledge graphs for automatic entity extraction and cross-modal relationship discovery for enhanced understanding.

multimodal GraphRAG is a framework designed by David Hughes and Amy Hodler to seamlessly integrate visual and textual data for more comprehensive insights and more accurate responses. It combines embeddings that capture visual and audio semantics, graph-based reasoning and explainable outcomes.

Multimodal for Knowledge Graphs (MM4KG) combines structured knowledge representations with deep learning techniques to handle diverse information sources.

In “Towards Multi-modal Graph Large Language Model“, researchers propose a unified framework of multi-modal graph data, task, and model, discovering the inherent multi-granularity and multi-scale characteristics in multi-modal graphs.

Graph databases grow and evolve

Graph databases are exploding, thanks to the AI boom,as Joe McKendrick writes on ZDNet. Graph databases are projected to have a five-year CAGR of 24% – 26% according to Gartner and the Business Research Company, respectively. The overall database market will grow 16% annually.

As AI and RAG have given a significant boost to both graph and vector databases, people are trying to establish how these two compare, and when and how to use each.

Andreas Blumauer compares vector and graph database semantics. Graphs don’t just store facts: they encode logic, preserve causality, and let you do symbolic + neural hybrid search, Shobhit Tankha chimes in. In André Lindenberg’s words: A database tells you what is connected. A knowledge graph tells you why.

Graph databases are bustling with activity. First, we saw the unveiling of not one, but two new vendors in the last couple of months. Tentris, an efficient disk-based graph database for RDF knowledge graphs, is now in open beta. And TuringDB, a low-latency in-memory graph database engine, is now open for early access.

Existing graph database vendors are making progress too.

Neo4j went HTAP by launching Infinigraph, a new graph architecture that aims to eliminate data silos between transactional and analytical systems.
Amazon Neptune 1.4.5 introduced engine improvements and support for AWS Graviton-based r8g instances.
Aerospike Graph Database 3.0 was announced, bringing improvements in developer ease, performance, and cost efficiency.
Graphwise announced the availability of versions 11 and 11.1 of GraphDB, bringing broad LLM compatibility, MCP support, precision entity linking, native GraphQL support and performance improvements.
Kuzu v0.11.0 was released, bringing single-file databases, improvements to vector and full-text search indices, and new LLM support.
TigerGraph announced a strategic investment from Cuadrilla Capital.

GQL, the newly-minted graph query language standard, is seeing adoption. Microsoft is adding GQL support to KQL graph semantics, enabling users to run GQL queries on any Fabric Eventhouse or Azure Data Explorer. Shortly after, Microsoft Fabric started offering graph analysis in Real-Time Intelligence. And Siren is the first investigative platform to offer GQL graph querying integrated with deep search.

For a guide to designing, querying, and managing graph databases with GQL, check the newly released book Getting Started with the Graph Query Language (GQL). And now you can directly run GQL queries in VS Code, thanks to the Ultipa VS Code Extensions.

Graph data models: LPG vs. RDF, OWL vs. SHACL

The LPG vs. RDF debate over graph data models never really goes away. Bryon Jacob explored RDF’s complete stack – identity (IRIs), structure (triples), semantics (RDFS/OWL), queries (SPARQL), and compared it to property graphs. Jacob argues that major enterprises are discovering they’ve been rebuilding RDF piece by piece.

Ora Lassila agrees, and Atanas Kiryakov has a go at debunking urban myths about RDF and explaining how ontologies help GraphRAG; good insights in the comment section. In bridging LPG and RDF frameworks, Enterprise Knowledge share ways to manage and apply a selection of these frameworks to meet enterprise needs.

In his exploration on the future of knowledge graphs, Kurt Cagle claims that we are soon likely to see the unification of LPG and RDF. Cagle also predicts that event-driven and dynamic knowledge graphs SHACL-based modeling will be more important going forward. He follows up arguing it’s time to rethink Linked Data.

Netflix unveiled its UDA (Unified Data Architecture) to model once, represent everywhere, enabling to transpile domain models into schema definition languages like GraphQL, Avro, SQL, RDF, and Java while preserving semantics. S&P launched its new AI-ready Metadata on the S&P Global Marketplace, with RDF under the hood.

People are also debating OWL vs. SHACL. Holger Knublauch, Boris Pelakh, Pete Rivett and Jessica Talisman address this in the great semantic modeling debate, while Michael Iantosca argues that both OWL and SHACL can be employed during the decision-making phase for AI Agents when using a knowledge graph.

Holger Knublauch shares a preview of what is coming in SHACL 1.2. Veronika Heimsbakk’s book “SHACL for the Practitioner” is open for pre-orders. And Kurt Cagle shows how to make pizza with AI and SHACL, and how to use SHACL to validate anything and build user interfaces.

Graph AI: GNNs, graph transformers and foundational models

We have already seen how to approach graph data models and databases coming from the relational world. But what is a ‘relational foundation model’?

The GenAI boom has given us powerful language models that can write, summarize and “reason” over vast amounts of text and other types of data. But these models don’t work for high-value predictive tasks like predicting customer churn or detecting fraud from structured, relational data.

Kumo’s approach, “relational deep learning,” promises to change that. Kumo’s relational foundation model generalizes the transformer architecture to automatically represent any relational database as a single, interconnected graph, and learns directly from this graph representation.

Kumo looks like the first to productize this. However, people in Google and Yandex are working on similar approaches too. Microsoft’s Janu Verma shares his own take on Graph Transformers. The evolution is far from over, and the future of graph AI promises to be even more deeply connected.

The Connected Data community is a great place for an introduction to Graph Learning and GNNs.The authors of the Geometric Deep Learning textbook have recently added a new chapter on graphs. Jure Leskovec shares what every data scientist should know about Graph Transformers and their impact on structured data.

Wrapping up from the world of graph AI with more new releases. PyG (PyTorch Geometric) is a library built upon PyTorch to easily write and train Graph Neural Networks for a wide range of applications related to structured data.

PyG has evolved significantly since its initial release, establishing itself as a leading framework for Graph Neural Networks. PyG 2.0 is a comprehensive update that introduces substantial improvements in scalability and real-world application capabilities.

GraGOD is a modern approach to time-series anomaly detection using GNN techniques. It is a PyTorch-based framework that provides a flexible and modular architecture for building and training GNN models for anomaly detection.

Graph science: Strong perfect graphs, the new Dijkstra’s algorithm and convergent neural networks

Last but not least, advances on the scientific front for graphs. Starting with a profile on Maria Chudnovsky, the “superstar mathematician” who solved a 40-year-old puzzle about perfect graphs. Chudnovsky’s breakthrough on the Strong Perfect Graph Conjecture shows how abstract math creates real-world solutions.

Speaking of real-world solutions: when we use Google Maps to find the fastest route, behind the scenes, it’s running some version of Dijkstra’s algorithm. That’s been the standard way to compute “shortest paths” since the 1950s. Researchers have found a faster way to run Dijkstra’s shortest path algorithm: something people thought couldn’t really be improved in a meaningful way.

Travis Thompson claims this transfers well into the way data products are built and consumed. Alexander Stage notes that this a great theory milestone, but production routing already “changed the rules” years ago with preprocessing and smart graph engineering.

Miklós Molnár reports on Szegedy Balázs’ work on training neural networks with identical architectures. Results point to the Platonic Representation Hypothesis, according to which neural networks are converging to a shared statistical model of reality. And Alberto Gonzalez shares the basics of representing graphs.

The post Graph is the new star schema. The Year of the Graph Newsletter Vol. 28, Autumn 2025 appeared first on The Year of the Graph.

The evolution of the graph technology and business landscape in 2025. Year of the Graph Newsletter V. 27, Spring–Summer 2025

George Anadiotis — Tue, 27 May 2025 11:02:31 +0000

Knowledge Graphs and Graph RAG Galore, New Graph Database Engines, Graph Analytics and Visualization, and Graph Foundation Models.

By George Anadiotis

Do you trust market research reports? If you do, here’s evidence #1 to consider: The global knowledge graph market is projected to reach $6.93 Billion by 2030 from $1.06 Billion in 2024, growing at a CAGR of 36.6%.

Do you trust analyst firms? If you do, here’s evidence #2 to consider: As many as 50% of Gartner client inquiries around the topic of AI involve a discussion around the use of graph technology.

Do you trust market signals? If you do, here’s evidence #3 to consider: graph-based products such as RDFox and data.world are powering household products such as the Samsung Galaxy S25 and ServiceNow, following their respective acquisitions.

All of these pieces of evidence point to the same direction: the graph landscape is evolving rapidly in terms of diversity, depth, and density and the outlook is positive, despite ups and downs.

But there’s good reason to read through this round of graph-related news and insights even if you don’t trust, or care about, any of the above. Read on to learn about how to build and visualize graphs, new graph database engines, variants on Graph RAG, a roadmap for graph analytics, and Graph Foundation Models, applications at scale, LLMs and graphs.

The Graph Technology Landscape
Knowledge Graph Market Outlook
Building and Evaluating Knowledge Graphs as Durable Assets
Knowledge Graphs Powering Household Names Through M&A
Knowledge Graphs as the Essential Truth Layer for Pragmatic AI
Graph RAG Galore
New Graph Database Engines, Standardization and Performance
Graph Analytics and Visualization: Roadmap, Features and Platforms
Graph Foundation Models, Applications at Scale, LLMs and Graphs

This issue of the Year of the Graph is brought to you by G.V(), metaphacts, Connected Data London, and Built to Last.

If you want to be featured in an upcoming issue and support this work, reach out!

You already understand the power of graph technology.

G.V() helps you understand your graph.

G.V() is a graph database IDE that helps you write, execute, and test queries; track your data model; explore and edit your graph data on the fly; and show your work with powerful graph data visualizations. Compatible with 18 different graph technologies and growing, G.V() is easy to use, low cost, low commitment, vendor agnostic, and plays well with any security architecture.

Try out G.V() for yourself and start querying your database in less than 5 minutes: gdotv.com

The graph technology landscape

The graph technology landscape infographic has been helping map the world of graph technology since 2014. Its goal is to introduce the key categories within the world of graph tech, and key players within those categories.

Of course, as its creators in Linkurious acknowledge, this is just a starting point, not a complete listing. It could not possibly be, for a domain that’s evolving so rapidly both in terms of R&D innovation as well as market growth.

Keeping track of the graph technology landscape requires constant attention and hard work, which is why the infographic and accompanying report are only updated once every 5 years.

The takeaway from the 2025 version is that more and more organizations are adopting graph technology, as they find it to be an asset for an ever increasing number of use cases. Bonus tracks: Graph technologies outlook in 2025 by Paco Nathan, predictions for knowledge graphs in 2025 by Tony Seale, and revisiting another graph tech landscape by Nicolas Figay.

AI you can trust, powered by semantics

When AI lacks enterprise specific context, it’s just guesswork. Coming July 2025, metis is metaphacts’ new knowledge-driven AI platform transforming disconnected enterprise data into real business value.

With metis, enterprises can:

● Ground AI responses in enterprise semantics for accuracy and trust

● Design and deploy custom knowledge-driven AI agents equipped with a

conversational interface

● Control and audit enterprise data usage

● Combine tools like summarization, entity linking and query execution—powered by

business-specific semantics

By grounding AI in semantics, metis delivers AI that truly understands your business, while ensuring security, explainability & trustworthiness. That’s the promise of knowledge-driven agentic AI—and it’s what powers metis.

Contact metaphacts to learn more!

Knowledge graph market outlook

The key drivers for the growing adoption of graph technology, as identified by Research and Markets’ Knowledge Graph Research Report 2025, are the rising demand for AI/generative AI solutions, the rapid growth in data volume and complexity, and the growing demand for semantic search.

The knowledge graph market is estimated at USD 1.06 billion in 2024 to USD 6.93 billion by 2030, at a Compound Annual Growth Rate (CAGR) of 36.6%. As Research and Markets notes, the Graph Database Engine segment is projected to hold the largest market size, and the services segment is projected to register the fastest growth rate during the forecast period.

Research and Markets notes lack of expertise and awareness as well as standardization and interoperability as the major challenges holding back growth in the market – although that’s changing. Lack of expertise also justifies the fact that services (consulting, presumably) is identified as the fastest growing part of the market.

Connected Data London is back!

The go-to conference for those who use the relationships, meaning, and context in Data to achieve great things.

Connecting Data, People & Ideas since 2016. Connected Data London provides a community, events, and thought leadership on all things Knowledge Graph, Graph Analytics, AI, Data Science, Graph Databases and Semantic Technology.

Whether you’re an engineer, data scientist, architect, or decision-maker, this is your chance to connect with the brightest minds shaping the future of Connected Data & the full program of:

Expert talks

Networking with innovators

Real-world case studies

Practical workshops

Community Dinner

Leonardo Royal Hotel Tower Bridge | November 20-21, 2025

Discounted Early bird tickets now available at 2025.connected-data.london

Building and evaluating knowledge graphs as durable assets

The Research and Markets report identifies “rapid proliferation of knowledge graphs” as an opportunity. While we certainly see signs of this, there are a few points worth making here.

As the popularity of knowledge graphs grows, their implementation complexity underscores the need to evaluate whether they are the right solution for the organization’s use case. This is what Gartner notes in its research note titled “How to Evaluate the Applicability of Knowledge Graphs for Your Use Cases“.

Gartner identified knowledge graphs at the heart of Critical Enabler technologies in its 2024 Hype Cycle for Emerging Technologies, and noted that “Adding Semantic Data Integration & Knowledge Graphs” was one of the Top 10 trends in Data Integration and Engineering for 2024.

However, Gartner finds that even though awareness of knowledge graph use cases is increasing, the willingness or business buy-in to invest in such initiatives is low. The benefits to the business still remain unclear, and organizations are still struggling to figure out when to use knowledge graphs to deliver business value.

As Mike Dillinger notes, rich knowledge graphs are durable assets – assets that have a long life span, providing utility or value over an extended period, usually in a business or economic context. They are typically not intended for sale but are absolutely essential for operations, like other assets such as property, equipment, and machinery. This is an extremely fitting metaphor.

Knowledge graphs are organizational CapEx, and they should be evaluated as such. Measuring their value should be based on what they enable, which ranges from data governance to AI applications.

In “Why Graph Implementations Fail“, Lulit Tesfaye notes that oftentimes, organizations have a limited understanding of the cost-benefit equation. Mel Richey addresses “What It Takes To Build a Great Graph“, and Gartner shares “How to Build Knowledge Graphs That Enable AI-Driven Enterprise Applications”.

Jessica Talisman chimes in with the Ontology Pipeline to power semantic knowledge systems, noting that a structured, scalable approach to semantic knowledge management can justify investments with well-defined ROI metrics and improve data quality and governance, essential for AI success. Jeremy Ravenel shares his experience on where to start when you want to build an ontology.

There’s no shortage of tools for building knowledge graphs either. Yassir Lairgi shared iText2KG, a Python package designed to incrementally construct consistent knowledge graphs with resolved entities and relations. WhyHow open sourced its Knowledge Graph Studio. And OpenSPG (Semantic-Enhanced Programmable Graph) is a new generation of enterprise knowledge graph (EKG) engine, bidirectionally enhanced by LLMs and knowledge graphs.

Knowledge graphs powering household names through mergers and acquisitions

The fact that knowledge graphs represent an investment for organizations who are serious about building AI is being increasingly understood. ServiceNow gets this, and is acquiring data.world to dig deeper into data for AI. As Joe Hilger notes, there is ongoing consolidation in the semantic software industry, driven by GenAI and semantic layers.

“According to a Gartner focus group, 4% of technology leaders believe that their data is AI-ready — that’s pretty sobering. Gartner goes on to say in a separate report that by 2026, 60% of AI projects will fail because the data is not AI-ready.” This is what Gaurav Rewari, senior vice president and general manager of data and analytics products at ServiceNow said in the context of data.world’s acquisition.

There are other data catalog specialists, but Rewari said Data.world’s knowledge graph implementation made it a good fit for ServiceNow. ServiceNow already has knowledge graph support, but data.world brings metadata collectors and knowledge graph expertise that can enrich ServiceNow’s graph further.

Earlier in the year, Samsung launched the new Galaxy S25 series with new AI features built on technology from Oxford Semantic Technologies. A spinout of Oxford University in 2017 by three of the world’s leading computer science professors in the field of knowledge-based AI technology, Oxford Semantic Technologies was acquired by Samsung Electronics in July 2024.

The company’s RDFox® technology is behind Samsung’s Personal Data Engine to create hyper-personalised user experiences leveraging knowledge graphs, and will be included in the latest Galaxy S25 series. Co-founder Ian Horrocks was among the speakers at Samsung’s Unpacked event.

Subscribe to the Year of the Graph Newsletter

Keeping track of all things Graph Year over Year

Knowledge graphs as the essential truth layer for Pragmatic AI

Organizations are facing a critical challenge to AI adoption: how to leverage their domain-specific knowledge to use AI in a way that delivers trustworthy results. Knowledge graphs can provide the missing “truth layer” for AI that transforms probabilistic outputs into real world business acceleration.

This is the takeaway from an in-depth conversation on knowledge graphs as the essential truth layer for Pragmatic AI with Tony Seale. It addresses everything from knowledge graph first principles to application patterns for safe, verifiable AI, real-world experience, trends, predictions, and the way forward. Seale, also known as “The Knowledge Graph Guy”, is the founder of the eponymous consulting firm.

Some related background material: “What are the Different Types of Graphs? The Most Common Misconceptions and Understanding Their Applications“, as well as “What is Semantics and Why Does it Matter?” – both by Enterprise Knowledge, highlighting different aspects of knowledge graph first principles. And Veronika Heimsbakk explains knowledge graphs for dummies – and beyond.

Graph RAG galore

Despite hurdles in GenAI adoption, or precisely because of it, making the most of GenAI-powered systems by combining them with trustworthy information in controlled environments, aka RAG (Retrieval Augmented Generation) has been consistently receiving attention. The Graph RAG variant, which we covered extensively in the previous YotG issue, is evolving and seeing its own variants.

For an introduction to Graph RAG, check out “Democratizing data with Graph RAG: What it is, What it can do, How to evaluate it“. Τhis led to the Connected Data Knowledge Graph challenge, and an open source implementation based on Kuzu and visualized on G.V(). A conceptual introduction to Graph RAG is also given by Jakob Pörschmann, who then details an implementation on the Google Cloud stack.

In the last few months, we have seen Microsoft open-sourcing its Graph RAG implementation, providing improvements such as auto-tuning for rapid adaptation to new domains and dynamic community selection, and releasing LazyGraphRAG. LazyGraphRAG is meant to address a key criticism of Graph RAG, namely that it’s costly to implement. But there are more issues with Graph RAG.

This is why people like Irina Adamchic come up with alternatives to address those. Adamchic introduced a three-layer fixed entity architecture for efficient RAG on graphs. It relies on leveraging domain knowledge in a so-called ontology layer. She also developed a NLP-based variant that doesn’t rely on domain knowledge, while Elena Kohlwey navigates the world of advanced RAG patterns.

OG-RAG, by contrast, is a RAG variant offering ontology-grounded retrieval-augmented generation for Large Language Models. KET-RAG promises 10x cheaper, smarter knowledge retrieval. MiniRAG introduces near-LLM accurate RAG for Small Language Models with just 25% of the storage. Mindful-RAG is a framework designed for intent-based and contextually aligned knowledge retrieval.

GFM-RAG is a graph foundation model for RAG. GNN-RAG uses a Graph Neural Retrieval for LLM reasoning. NodeRAG is a Graph RAG variant which uses heterogeneous graphs with fine-grained semantic units, entities, relationships, and high-level summaries instead of homogeneous graphs. SimGRAG transforms queries into graph patterns and aligns them with candidate subgraphs using a graph semantic distance metric.

mmGraphRAG taps into non-textual data like images and audio. PathRAG effectively reduces redundant information with flow-based pruning, while guiding LLMs to generate more logical and coherent responses with path-based prompting. CDF-RAG iteratively refines queries, retrieves structured causal graphs, and enables multi-hop causal reasoning across interconnected knowledge sources.

A survey on Graph RAG formalizes the GraphRAG workflow, outlines the core technologies and training methods at each stage, examines downstream tasks, application domains, evaluation methodologies, and industrial use cases and explores future research directions.

Paco Nathan unbundles the Graph in GraphRAG, Lettria, Jay Yu and May Habib benchmark it, and Francois Vanderseypen shares a conceptual overview for building robust solutions and a collection of graph RAG open source stacks to generate and visualize knowledge graphs.

Eventually, it may be Hybrid RAG approaches that may prove flexible enough to adjust to different scenarios. For many applications, a combination of retrieval methods, orchestrated by a smart router, may provide the best balance of performance and flexibility.

New graph database engines, standardization and performance

Although Graph RAG has been a boon for graph database vendors, it’s also possible to implement Graph RAG without a graph database. Aggregating graph databases and knowledge graphs in one category as the Research and Markets report did may not be something everyone will agree with. But it’s understandable why graph databases are marked as the largest part of this aggregated category.

Graph databases have been around for a long time by now. Like all SQL antagonists, their utility is sometimes questioned, and they have ups and downs in terms of mindshare. But overall they are on a growth trajectory, with use cases being increasingly understood, standardization and educational resources facilitating adoption, and important new developments in the market.

In “Graph Databases after 15 Years – Where Are They Headed?“, Gábor Szárnyas summarizes the history of graph database systems, focusing on their main categories and use cases. He then discusses the key challenges that continue to hinder the adoption of graph databases, including a fragmented landscape and performance limitations.

He wraps up with recent positive developments: (1) Advances in standardization that led to the ISO GQL and SQL/PGQ languages, (2) Performance improvements, (3) A new generation of open-source graph database systems. Bonus tracks – Semih Salihoğlu’sintroduction and background on graph databases, and Joe Dreyer’s guide to graph databases.

We have covered the developments in standardization in previous newsletters. Keith Hare shares current status, and Alastair Green highlights new GQL features, and the ongoing effort to close the “ontology gap” and align RDF vocabularies and LPG schemas. Alex Milowski also elaborates on GQL Schemas and Types.

Graph database vendors keep evolving as well. Google entered the graph database market with Spanner Graph, AWS took one more step towards the One Graph vision for Neptune, and Neo4j released new self-service and GenAI features. Spanner Graph is now GA, Neo4j has introduced Aura Graph Analytics, and Amazon Neptune powers Amazon Bedrock Knowledge Bases with Graph RAG.

At the same time, we also notice the evolution in graph database performance, with new vendors and engines such as Neo4j BIFROST, Stardog BARQ, Aerospike Graph, Data Graphs, QLever, Kuzu, and HugeGraph.

What if fitness wasn’t about following a strict routine—but about building a plan that works for YOU?

“Built To Last” isn’t just another fitness book. It’s a practical, customizable guide designed to help you create a sustainable approach to health, movement, and longevity. Whether you’re just starting your journey or fine-tuning your current routine, this book gives you the tools to take control of your fitness—on your terms.

Why Built To Last is Different

Unlike other fitness books that push quick fixes and extreme regimens, Built To Last is based on real science, real results, and real sustainability.

Graph analytics and visualization: roadmap, features and platforms

How might forthcoming graph processing systems deliver extensive scalability, efficiency, and versatile querying and analytical functionalities to meet the diverse demands of real-world scenarios? This is the overarching question that a panel of experts set out to address in a ACM Sigmod panel. They published their findings in “A Roadmap to Graph Analytics“.

Visual analytics is a natural complement for graph analytics. The i2 Group published a list of considerations for knowledge graph visualization and analytics, highlighting Flexibility, Ontology Alignment , Visualization, Dynamic Styling, Aggregation, Grouping and Summarization, Customization, Semantics, De-duplication and Entity Resolution, Data Lineage and support for Special types.

New tools and capabilities for graph visualization were unveiled recently. G.V() is a graph database client & visualization tool that started out with Gremlin and now also supports Neo4j, Memgraph, and Neptune Analytics. RDF-to-Gephi is an open source tool to visualize RDF knowledge graphs. And a new yFiles open source widget makes it convenient to add graph visualizations of SPARQL queries to Jupyter Notebooks.

Graph Foundation Models, applications at scale, LLMs and graphs

We also have new developments In the area of Graph AI, some of them sparked by the LLM wave, some continuing the existing wave of innovation. Michael Galkin argues that the era of Graph Foundation Models has begun, and provides a few examples of how one can use them already today. Case in point: AnyGraph, a Graph Foundation Model in the wild.

Galkin also warns that Graph Learning may lose relevance due to poor benchmarks, while a group of researchers from Huawei and research institutes explore the foundations and frontiers of Graph Learning theory. If you want to get up to speed with GNNs, this introduction to Graph Neural Networks and this list of must-read papers on GNNs will be handy.

Researchers from Amazon introduce GraphStorm, an all-in-one open source graph machine learning framework for industry applications that has been used and deployed for over a dozen billion-scale industry applications. Snapchat also uses large-scale graph neural networks in production, leveraging their own open source framework called GiGL (Gigantic Graph Learning).

Google’s Bryan Perozzi shares how graphs can help represent structured data for LLMs, covering graph encoding, GraphTokens, Transformer graph reasoning, and using graphs for synthetic data generation. Also of note on the latter topic, RDFGraphGen and LinkGen.

In “Plan Like a Graph”, researchers improve LLM task execution by decomposing them into sub-tasks that can be either solved by the LLM in parallel or sequentially. All these sub-tasks form an execution graph. GraCoRe is a benchmark for graph comprehension and complex reasoning in LLMs.

Bryan Perozzi also shared his team’s work on Graph reasoning with LLMs. And “A Survey of Large Language Models for Graphs” introduces a novel taxonomy for categorizing existing methods that combine LLMs and GNNs.

A different way to mix graphs and Language models: Graph Language Models are graph transformers, which enables graph reasoning. Simultaneously, they inherit and exploit Language Model weights, enabling them to represent and contextualize triplets in a Graph of Triplets. Relational Graph Transformers is a new architecture transforming how we extract intelligence from relational databases.

Wrapping up with a thought – provoking question, and a new way to look at LLMs: What if your LLM is a graph? As Pierre-Carl Langlais comments on the analysis shared by Petar Veličković, once you start seeing LLMs as graph neural networks, many structural oddities suddenly fall into place. In what is arguably the flip side, Kurt Cagle lays out the differences between LLMs and knowledge graphs.

This issue of the Year of the Graph is brought to you by G.V(), metaphacts, Connected Data London, and Built to Last.

If you want to be featured in an upcoming issue and support this work, reach out!

The post The evolution of the graph technology and business landscape in 2025. The Year of the Graph Newsletter Vol. 27, Spring – Summer 2025 appeared first on The Year of the Graph.

Graph Database market update, September 2024: Google Cloud Spanner Graph, Amazon Neptune, Neo4j

George Anadiotis — Wed, 04 Sep 2024 13:07:46 +0000

Google enters the Graph Database market with Spanner Graph, AWS takes one more step towards the One Graph vision for Neptune, and Neo4j releases new self-service and GenAI features.

By George Anadiotis

There is no news in August. While you should not take this Umberto Eco quote 100% to heart, the truth is that major news are rarely announced in August. PR people are people too, and who likes to work in the month on which traditionally most people take their much needed time off – which also means going offline?

Product news are no different, and making product announcements in August seems like an odd choice. Nevertheless, this is what both AWS and Google just did in August 2024. Both announcements have to do with their Graph Database offerings, moving them forward in significant ways. Neo4j on the other hand waited until September 4. Here is a breakdown of the news and its impact on the market.

The Big 3 cloud providers – AWS, Microsoft Azure and Google Cloud – collectively make up for the vast majority of the cloud market. Even with if the rumored return to on-premises is going to materialize, the Big 3 still exert a huge influence over technology directions and budgets. When one of the Big 3 enters a new market, that immediately does a number of things.

First, it gives a boost to the market, as it provides validation and exposes it to a wider audience. Second, it forces everyone already in the market to take note and up their game, for fear of having their lunch eaten. Third, it creates co-opetition dynamics, as practically all vendors have to work with the Big 3 to make their products available to their clients.

We have seen this dynamics at play already in 2017 and 2018, when Microsoft and AWS made their Graph Database offerings available: Cosmos DB and Neptune, respectively. In August 2024, Google got into the Graph database action with Cloud Spanner Graph and AWS took another step towards graph data model integration. Neo4j on its part announced new capabilities for Aura, its cloud offering.

Why now, how to evaluate those offerings, and what does that all mean for the Graph Database market? These are the questions this brief analysis is meant to address.

Google Cloud Spanner goes Graph

Google Cloud Spanner is a globally distributed operational database that’s been generally available since 2017. The in-house version of Spanner was originally built by Google to handle workloads like AdWords and Google Play, that were, according to Google, previously running on massive, manually sharded MySQL implementations.

As Andrew Brust explains in his 2017 introduction of Spanner for ZDNet, Google needed a database that had native, flexible sharding capabilities, adhered to relational schema and storage, was ACID-compliant and supported zero downtime. Since such a database didn’t exist, Google created its own, and thus Spanner was born. After battle-testing the product in-house, Google made Cloud Spanner generally available.

In August 2024, Google introduced what it dubbed Spanner Graph, adding graph capabilities to Spanner. Spanner now supports the newly introduced ISO Graph Query Language (GQL). Spanner takes an idiosyncratic approach, effectively mixing the relational and graph model. It also has built-in search capabilities, and offers scalability, availability, and consistency as well as integration with Vertex AI for AI-powered insights.

For the most part, these features leverage Google’s ecosystem (Vertex AI and search) as well as the pre-existing capabilities of Spanner (scalability, availability, and consistency). But the core of the announcement really lies in the way graph features have been implemented on top of the existing SQL interface.

Google Cloud Spanner Graph lets users mix and match relational and graph data models

What this implies is that Spanner has not really implemented a new engine for graph. Instead, Spanner seems to have opted for adding a GQL interface on top of its existing SQL engine, as well as a way to designate which tables correspond to nodes and edges in the graph. Using the GQL interface, users are able to execute graph queries on Spanner, and mix them with SQL queries as well.

Is that a good thing? And how well that will it work in practice? Spanner Graph seems like an implicit nod to Stonebraker and Pavlo, who argue something along the lines of “SQL is all you need”. It’s certainly possible to create a graph database interface on top of a relational database, and others have done it previously too – Microsoft SQL Server Graph and Oracle Graph come to mind.

However, working with graphs is not as simple as adding a few tables to model graphs using a relational engine. First, because using SQL to express graph queries would result in extremely complicated queries that are a nightmare to create and maintain. And second, because running these queries on the underlying relational engine can lead to a huge amount of joins, having a dramatic effect on performance.

How does Spanner Graph work?

Spanner has addressed the first point by implementing GQL. That enables users to express queries in a bona fide graph query language, making Google one of the first vendors to offer support for the newly minted GQL standard. The fact that GQL and SQL can be mixed and matched when writing queries is an interesting feature. If nothing else, it hints at how Spanner works under the hood.

To address the second point, however, GQL itself is not enough. Simply translating GQL to SQL and running it over relational tables could lead to problematic performance for Spanner Graph. At this point, we don’t really know whether Spanner Graph uses graph-specific optimizations. What we do know is that Spanner claims to offer single-digit ms latency for SQL queries, and that Spanner’s distributed nature calls for attention when executing those SQL queries.

We don’t currently have any 3rd party references regarding Spanner Graph’s performance. Except for a quote provided by the Credit Karma Engineering team, we don’t know how Spanner Graph is used in production either. We hope that more material will be made available to clarify those points, crucial to inform purchase decisions in the Graph Database market.

In any case Spanner Graph makes for an interesting offering for existing Spanner users, which presumably is Google’s initial primary target audience. Google lists a number of use cases for Spanner Graph: Product recommendations, Financial fraud detection, Social networks, Gaming, Network security, and Graph RAG.

All of the above are indeed prime examples of use cases for Graph Databases. And it’s been that way for a long time – with the obvious exception of Graph RAG. Graph RAG may well be what tipped the balance to make Google decide to address the Graph Database gap in its portfolio by adding graph features to Spanner. Presumably, Google wants a piece of the GenAI action, and checking the Graph RAG box could help.

Querying RDF knowledge graphs with openCypher on Amazon Neptune

It may seem like Google decided to come out and play Graph out of the blue. But as reported by YotG reader Seb Heymann, Bei Li, Spanner Graph cofounder, has contributed to the GQL ISO group for the last 5 years. That’s long before GenAI and Graph RAG were a thing.

AWS, by contrast, has been a force to be reckoned with in the Graph Database market since Neptune’s debut in 2018. Its latest announcement is that Neptune users can now build and deploy knowledge graphs with RDF and openCypher.

As Neptune’s team shares, when users build an application that uses a Graph Database, they’re typically faced with a technology choice at the start: There are two different types of graphs, Resource Description Framework (RDF) graphs and labeled property graphs (LPGs), and their choice of which to use will determine which query languages they can use.

RDF graphs use SPARQL as their query language, and for LPGs the query languages are Gremlin and openCypher. Users and developers who are new to graph technology are often confused about having to make this choice, which Neptune’s team calls “a rift in the entire graph industry”. Reasons for this division are manifold: technology limitations, lack of awareness, and strongly held views.

RDF, with its standardized serialization formats, global identifiers, and the availability of Linked Open Data sets, is of particular value for data architects who seek to build, integrate, and interchange graph data. Application development teams often prefer LPG query languages such as openCypher, due to their intuitive syntax, ecosystem maturity, and features such as built-in support for path extraction and algorithms.

Amazon Neptune lets users use RDF knowledge graphs with the openCypher query language

Neptune’s team has been working for a long time towards what they call the One Graph vision, so this move is by no means a surprise. It’s the next logical step towards the implementation of that vision. Neptune has a lot to gain by this, as the end game may even be unifying the two engines they currently operate for Neptune under the hood – one for RDF and one for LPG data.

For Neptune, that would mean having a much cleaner and simpler implementation with less overhead and more efficiency. For users, being graph data model agnostic should make life simpler and easier and offer more choice.

One caveat, however, is that mixing and matching data models and query languages may not be entirely straightforward as an interim state, especially for beginners. Either way, we expect to see Neptune continuing the work towards the realization of the One Graph vision, with the next stop probably being support for GQL.

Neo4j gets Aura GenAI, self-service, performance and security features

In contrast with Google and AWS, Neo4j focuses all its energy on one thing only – its eponymous Graph Database product. This, in fact, is what its CEO Emil Eifrem has long ago called out as Neo4j’s competitive advantage against the Googles and AWSs of the world.

Like AWS, Neo4j is also walking a consistent path in terms of its product roadmap. Sudhir Hasbe, the CPO Neo4j appointed in 2023, set the tone immediately. Neo4j has gone all in on GenAI and Large Language Models, with integrations and related features having been released ever since.

GenAI may be a relatively new priority for Neo4j, but Aura, Neo4j’s fully managed cloud, is a long-standing one. Not only have GenAI features been deployed with a priority on Aura, but now it’s Aura’s turn to be given a GenAI lift. Neo4j just announced an array of new features as well.

Aura gets a new console with GenAI co-pilot as a unifying hub to easily administer, manage, ingest, model, and visualize data efficiently across all of Neo4j’s offerings and tools. There is also a new self-serve offering, Neo4j AuraDB Business Critical, at a 20%+ lower price point than Neo4j’s traditional AuraDB Enterprise offering and designed for highly available workloads requiring advanced security.

Another new feature is the new no/low-code interactive dashboard builder NeoDash that lets users create maps, graphs, bar and line charts, tables, and other visuals for anyone to understand, analyze, and interact. While developers have used NeoDash as part of Neo4j Labs, it is now a fully supported offering for enterprises.

Last but not least, there are performance and security features announced. 15x scale improvement in real-time read capacity, and advanced enterprise control, audit, and compliance capabilities that include Customer Managed Encryption Keys and Security Log Forwarding.

The Graph Database market is moving forward

Compared to the announcements from Google and AWS, it would be fair to say that Neo4j’s features seems like incremental improvements. In all fairness, however, Neo4j has been consistent in releasing such features. This signifies the constant evolution of Neo4j’s offering.

Overall, the news paints an interesting picture. Google seems to be operating in typical Big 3 fashion, ever-expanding its offering. AWS has a dedicated product team for its Neptune database, and the same obviously holds for pure-play Graph Database vendors, with Neo4j exemplifying this.

The one vendor that’s practically missing from this picture is Microsoft. In theory, Microsoft offers Graph Database capabilities for Cosmos DB, an offering similar in some ways to Spanner, as well as SQL Server, its relational database timeless hit. In practice, however, both seem stagnant and not actively developed graph-wise.

It would be interesting to see whether Microsoft, whose work on Graph RAG greatly benefits Graph Database vendors, decides to actively pursue this market too. Either way, the market is moving forward, and we’ll be keeping an eye out to evaluate how these news move the needle.

The post Graph Database market update, September 2024: Google Cloud Spanner Graph, Amazon Neptune, Neo4j appeared first on The Year of the Graph.

Knowledge Graph Enlightenment, AI and RAG. The Year of the Graph Newsletter Vol. 26, Summer 2024

George Anadiotis — Wed, 26 Jun 2024 15:29:15 +0000

A snapshot of the adoption wave for graphs in the real world, and the evolution of their use to support and advance AI – generative or otherwise. Is Knowledge Graph enlightenment here, and what does that mean for AI and RAG?

In the previous edition of the YotG newsletter, the wave of Generative AI hype was probably at its all-time high. Today, while Generative AI is still talked about and trialed, the hype is subsiding. Skepticism is settling in, and for good reason. Reports from the field show that only a handful of deployments are successful.

At its current state, Generative AI can be useful in certain scenarios, but it’s far from being the be-all and end-all that was promised or imagined. The cost and expertise required to evaluate, develop and deploy Generative AI-powered applications remains substantial.

Promises of breakthroughs remain mostly promises. Adoption even by the likes of Google and Apple seems haphazard with half-baked announcements and demos. At the same time, shortcomings are becoming more evident and understood. This is the typical hype cycle evolution, with Generative AI about to take a plunge in the trough of disillusionment.

Ironically, it is these shortcomings that have been fueling renewed interest in graphs. More specifically, Knowledge Graphs, as part of RAG (Retrieval Augmented Generation). Knowledge Graphs are able to deterministically deliver benefits.

Having preceded Generative AI by many years, Knowledge Graphs are entering a more productive phase in terms of their perception and use. Coupled with proper tools and oversight, Generative AI can boost the creation and maintenance of Knowledge Graphs.

Knowledge Graphs as Critical Enablers reaching the Slope of Enlightenment

Gartner’s Emerging Tech Impact Radar highlights the technologies and trends with the greatest potential to disrupt a broad cross-section of markets. Gartner recently published a list of 30 emerging technologies identified as critical for product leaders to evaluate as part of their competitive strategy.

Knowledge Graphs are at the heart of Critical Enabler technologies. This theme centers on expectations for emerging applications — some of which will enable new use cases and others that will enhance existing experiences — to guide which technologies to evaluate and where to invest.

The Gartner Emerging Tech Impact Radar highlights the technologies and trends with the greatest potential to disrupt a broad cross-section of markets

A few days later, at Gartner D&A London, “Adding Semantic Data Integration & Knowledge Graphs” was identified as one of the Top 10 trends in Data Integration and Engineering.

And just a few days before this newsletter issue came out, the Gartner 2024 Hype Cycle for Artificial Intelligence was released. As Research VP, AI at Gartner Svetlana Sicular notes, investment in AI has reached a new high with a focus on generative AI, which, in most cases, has yet to deliver its anticipated business value.

This is why Gen AI is on the downward slope on the Trough of Disillusionment. By contrast, Knowledge Graphs were there in the previous AI Hype Cycle, and have now moved to the Slope of Enlightenment.

Graph RAG: Approaches and Evaluation

It was only 6 months ago when people were still exploring the idea of using knowledge graphs to power RAG. Even though people were using the term Graph RAG before, it was the eponymous publication by a research team in Microsoft that set the tone and made Graph RAG mainstream.

Since the beginning of 2024, there have been 341 arXiv publications on RAG, and counting. Many of these publications refer to Graph RAG, either by introducing new approaches or by evaluating existing ones. And that’s not counting all the non-arXiv literature on the topic. Here is a brief list, and some analysis based on what we know so far.

In “GraphRAG: Design Patterns, Challenges, Recommendations”, Ben Lorica and Prashanth Rao explore options based on their experience both on the drawing board and in the field. In “GNN-RAG: Graph Neural Retrieval for Large Language Model Reasoning”, C. Mavromatis and G. Karypis introduce a novel method for combining LLMs with GNNs.

Terence Lucas Yap runs the course “From Conventional RAG to Graph RAG”. Both Neo4j and LlamaIndex have been independently working on Graph RAG, until eventually they joined forces as LlamaIndex introduced the Property Graph Index. LinkedIn shared how leveraging a Graph RAG approach enabled cutting customer support resolution time by 29.6%.

Chia Jeng Yang wrote about “The RAG Stack: Featuring Knowledge Graphs”, highlighting that as attention shifts to a ‘RAG stack’, knowledge graphs will be a key unlock for more complex RAG and better performance. Daniel Selman has been researching and building a framework that combines the power of Large Language Models for text parsing and transformation with the precision of structured data queries over Knowledge Graphs for explainable data retrieval.

GraphRAG: Unlocking LLM discovery on narrative private data

Graph RAG has not been around for long, but multiple efforts at evaluation are already under way. In “Chat with Your Graph” Xiaoxin He et.al introduce G-Retriever, a flexible graph question-answering framework, as well as GraphQA, a benchmark for Graph Question Answering. “A Survey on Retrieval-Augmented Text Generation for Large Language Models” by Huang and Huang presents a framework for evaluating RAG, in which SURGE, a Graph-based method, stands out.

Writer compared Knowledge Graph with other RAG approaches on the basis of accuracy, finding that Knowledge Graph achieved an impressive 86.31% on the RobustQA benchmark, significantly outperforming the competition. Sequeda and Allemang did a follow-up to their previous evaluation, finding that utilizing an ontology reduces the overall error rate to 20%.

In Jay Yu’s micro-benchmark on the performance of GraphRAG, Advanced RAG and ChatGPT-4o, findings were more nuanced. GraphRAG started strong but stumbled due to its knowledge graph dependency. ChatGPT-4o was a general knowledge champ, but it missed a couple of questions. Advanced RAG’s modular architecture clinched the win.

For LinkedIn, RAG + Knowledge Graphs cut customer support resolution time by 28.6%. LinkedIn introduced a novel customer service question-answering method that amalgamates RAG with a knowledge graph. This method constructs a knowledge graph from historical issues for use in retrieval, retaining the intra-issue structure and inter-issue relations.

As Xin Luna Dong shared in her SIGMOD Keynote “The Journey to a Knowledgeable Assistant with Retrieval-Augmented Generation (RAG)”, there are some clear takeaways. Good metrics are key to quality. Knowledge Graphs increase accuracy and reduce latency, although reducing latency requires relentless optimization. Easy tasks can be distilled to a small LM, and summarization plays a critical role in reducing hallucinations.

For a deeper dive, there’s a book by Tomaž Bratanič and Oskar Hane: Knowledge Graph-Enhanced RAG, currently in Manning Early Access Program (MEAP), set for publication in September 2024.

Connected Data London: Bringing together Leaders and Innovators

Jay Yu has also released a number of chatbots in the last few months, based on the writings of graph influencers such as Kurt Cagle, Mike Dillinger, and Tony Seale and leveraging LLMs and RAG. There is something else Kurt, Mike and Tony all have in common too: they will be part of the upcoming Connected Data London 2024 conference.

Connected Data is back in London, for what promises to be the biggest, finest and most diverse in the Connected Data events to date. Join in the City of London on December 11-13 at etc Venues St. Paul’s for a tour de force in all things Knowledge Graph, Graph Analytics / AI / Data Science / Databases and Semantic Technology.

Submissions are open across 4 areas: Presentations, Masterclasses, Workshops and Unconference sessions. There is also an open call for volunteers and sponsors.

If you are interested in learning more and joining the event, or just want to learn from the experts comprising Connected Data London’s Program Committee as they explore this space, mark your calendars.

Connected Data London is organizing a Program Committee Roundtable on July 3, at 3pm GMT. More details and registration link here.

Advances in Graph AI and GNN Libraries

There are many advances to report on in the field of Graph AI / Machine Learning / Neural Networks. The best place to start would be to recap progress made in 2023, which is what Michael Galkin and Michael Bronstein do. Their overview in 2 parts covers Theory & Architectures and Applications.

But there is lots of ongoing and future work as well. In terms of research, Azmine Toushik Wasi compiled a comprehensive collection of ~250 graph and/or GNN papers accepted at the International Conference on Machine Learning 2024.

And it’s not just theory. LiGNN is a large-scale Graph Neural Networks (GNNs) Framework developed and deployed at LinkedIn, which resulted in improvements of 1% of Job application hearing back rate and 2% Ads CTR lift. Google has also been working on a number of directions. Recently Bryan Perozzi summarized these ideas in “Giving a Voice to Your Graph: Representing Structured Data for LLMs”.

Graph & Geometric ML in 2024: Where We Are and What’s Next

As far as future directions go, Morris et.al argue that the graph machine learning community needs to shift its attention to developing a balanced theory of graph machine learning, focusing on a more thorough understanding of the interplay of expressive power, generalization, and optimization.

Somewhere between past, present and future, Michael Galkin and Michael Bronstein take a stab at defining Graph Foundation Models, keeping track of their progress and outlining open questions. Galkin, Bronstein at.al present a thorough review of this emerging field. See also GFM 2024 – The WebConf Workshop on Graph Foundation Models.

If all this whetted your appetite for applying these ideas, there are some GNN libraries around to help, and they have all been evolving.

DGL is framework agnostic, efficient and scalable, and has a diverse ecosystem. Recently, version 2.1 was released featuring GPU acceleration for GNN data pipelines.
MLX-graphs is a library for GNNs built upon Apple’s MLX, offering fast GNN training and inference, scalability and multi-device support.
PyG v2.5 was released featuring distributed GNN training, graph tensor representation, RecSys support, PyTorch 2.2 and native compilation support.

Last but not least in the chain of bringing Graph AI to the real world, NVIDIA introduced WholeGraph Storage, optimizing memory and retrieval for Graph Neural Networks, and extended its focus to its role as both a storage library and a facilitator of GNN tasks.

Graph Database Market Growth and the GQL standard

Gartner analysts Adam Ronthal and Robin Schumacher, Ph.D. recently published their market analysis, including an infographic stack ranking of revenue in the DBMS market. This is a valuable addition to existing market analysis, as it covers what other sources typically lack: market share approximation.

The analysis includes both pure-play graph database vendors (Neo4j and TigerGraph), as well as vendors whose offering also includes graph (AWS, Microsoft, Oracle, DataStax, AeroSpike, and Redis – although its graph module was discontinued in 2023).

The dynamics at the top, middle and bottom of the stack are pretty much self-explanatory, and Neo4j and TigerGraph are on the rise. A propos, Neo4j keeps on executing its partnership strategy, having just solidified the partnerships with Microsoft and Snowflake.

It would also be interesting to explore how much graph is contributing to the growth of other vendors, but as Ronthal notes, the granularity of the data does not enable this.

GQL, the new standard in Graph query languages, is officially announced by the ISO

In other Graph DB news, Aerospike announced $109M in growth capital from Sumeru Equity Partners. As per the press release, the capital injection reflects the company’s strong business momentum and rising AI demand for vector and graph databases. Note the emphasis on Graph, coming from a vendor that is a recent entry in this market.

Another new entry in the Graph DB market is Falkor DB. In a way, Falkor picks up from where Redis left off, as it’s developed as a Redis module. Falkor is open source and supports distribution and the openCypher query language. It’s focused on performance and scalability, and targets RAG use cases.

Speaking of query languages, however, perhaps the biggest Graph DB news in a while is the official release of GQL. GQL (Graph Query Language) is now an ISO standard just like SQL. It’s also the first new ISO database language since 1987 — when the first version of SQL was released. This will help interoperability and adoption for graph technologies.

For people who have been involved in this effort that started in 2019, this may be the culmination of a long journey. Now it’s up to vendors to implement GQL. Neo4j has announced a path from openCypher to GQL, and TigerGraph also hailed GQL. It’s still early days, but people are already exploring and creating open source tools for GQL.

Knowledge Graph Research, Use Cases and Data Models

Wrapping up this issue of the newsletter with more Knowledge Graph research and use cases. In “RAG, Context and Knowledge Graphs” Kurt Cagle elaborates on the tug of war between machine learning and symbolic AI, manifested in the context vs. RAG debate. As he notes, both approaches have their strengths as well as their issues.

In “How to Implement Knowledge Graphs and Large Language Models (LLMs) Together at the Enterprise Level”, Steve Hedden surveys current methods of integration. At the same time, organizations such as Amazon, DoorDash and the Nobel Prize Outreach share how they did it.

There are also many approaches for creating Knowledge Graphs assisted by LLMs. Graph Maker, Docs2KG and PyGraft are just a couple of these. This almost begs the question – can Knowledge Graph creation be entirely automated? Are we looking at a future in which the job of Knowledge Graph builders, aka ontologists, will be obsolete?

The answer, as is most likely for most other jobs too, is probably no. As Kurt Cagle elaborates in “The Role of the Ontologist in the Age of LLMs”, an ontology, when you get right down to it, can be thought of as the components of a language.

LLMs can mimic and recombine language, sometimes in a seemingly brilliant and creative way, but they don’t really understand either language, or the domain it’s used to describe. They may be able to produce a usable model, but the knowledge and effort needed to verify, debug and complement it are not negligible.

As Cagle also notes, some ontologies may have thousands of classes and hundreds of thousands of relationships. Others, however, are tiny, with perhaps a dozen classes and relationships, usually handling very specialized tasks.

Cagle mentions SKOS, RDFS, and SHACL as examples of small ontologies handling specialized tasks. What they all handle is ontology, or more broadly, model creation itself. The art of creating ontological models for knowledge graphs, as Mike Dillinger points out, often starts with taxonomies.

Enhancing Knowledge Graphs with LLMs

Taxonomies – coherent collections of facts with taxonomic relations – play a crucial and growing role in how we – and AIs – structure and index knowledge. Taken in the context of an “anatomy” of knowledge, taxonomic relations – like instanceOf and subcategoryOf – form the skeleton, a sketchy, incomplete rendering of a domain.

Still, taxonomies are the structural core of ontologies and knowledge graphs as well as the foundation of all of our efforts to organize explicit knowledge. Dillinger believes that we can do better than today’s taxonomies – what he calls Taxonomy 2.0. He shares his take on building knowledge graphs in “Knowledge Graphs and Layers of Value”, a 3-part series.

Building these semantic models may be slow, as Ahren Lehnert notes in “The Taxonomy Tortoise and the ML Hare”. But it enables fast-moving machine learning models and LLMs to be grounded in organizational truths, allowing for expansion, augmentation, and question-answering at a much faster pace but backed with foundational truths.

All of the above point to semantic knowledge graphs and RDF. When it comes to choosing the right type of graph model, the decision typically boils down to two major contenders: Resource Description Framework (RDF) and Labelled Property Graphs (LPG).

Each has its own unique strengths, use cases, and challenges. In this episode of the GraphGeeks podcast hosted by Amy Hodler, Jesús Barrasa and Dave Bechberger discuss how these approaches are different, how they are similar, and how and when to use each.

GQL, mentioned earlier, applies to LPG. But it could also be used as a means to bring the two worlds closer together. This is what Ora Lassila explores in his “Schema language for both RDF and LPGs” presentation, also building on his previous work with RDF and reification. Semih Salihoğlu and Ivo Velitchkov both praise RDF, listing pros and cons and seeing it as an enabler for liberating cohesion, respectively.

The post Knowledge Graph Enlightenment, AI and RAG. The Year of the Graph Newsletter Vol. 26, Summer 2024 appeared first on The Year of the Graph.

Graphs, analytics, and Generative AI. The Year of the Graph Newsletter Vol. 25, Winter 2023 – 2024

George Anadiotis — Tue, 05 Dec 2023 08:39:25 +0000

Graphs, analytics and Generative AI. An account of the different ways graphs and AI mingle, plus industry and research news.

Is a generative AI preamble necessary for a newsletter focused on Knowledge Graphs, Graph Databases, Graph Analytics and Graph AI? Normally, it should not be. However, the influence of generative AI on the items included in this issue was overwhelming. There is a simple explanation for that.

It’s been a year since Generative AI burst into the mainstream with the release of ChatGPT. Notwithstanding a rather spotty record both in terms of technical performance and accuracy as well as in terms of business reliability, there’s no denying that Generative AI has captured the attention of executives worldwide.

Mentions of “generative AI” on earnings calls have skyrocketed since ChatGPT made its debut, rising from 28 in Q4’22 to 2,081 in Q3’23 — a 74x increase. With the majority of companies being at an early stage in their AI journey, executives feel the pressure to act amid the generative AI rush.

What this means is that generative AI has tremendous mindshare. Adoption is expected to grow 36% YoY 2030 according to Forrester, and more than 100M people in the US alone will use generative AI next year. Therefore, vendors are looking to position their products accordingly.

Played right, this can be more than a marketing scheme. There are ways in which graphs and generative AI can complement each other, leading to responsible enterprise decisions with knowledge-enriched Generative AI.

Let’s take a brief tour of product offerings and research efforts in that direction unveiled in the last few months.

Vectors, graphs, and RAG

RAG stands for Retrieval Augmented Generation. It’s a technique through which Large Language Models like ChatGPT can be instructed to use specific background knowledge to contextualize their processing. This offers a conversational interface to proprietary data, making LLMs applicable to business scenarios that require this.

RAG is the main reason why the interest in vector databases has skyrocketed in the last year. Like all machine learning models, LLMs work with vectors. Therefore, having a vector database to store information and feed it to LLMs seems like a reasonable choice for RAG. However, it’s not the only one. In fact, as people like Damien Benveniste argue, Graph Databases may be a better choice for RAG.

Vector Database Vs. Graph Database for RAG

Using graphs, the relationships between the different entities in the text can be extracted and used to construct a knowledge base of the information contained within the text. An LLM is good at extracting that kind of triplet information: [ENTITY A] -> [RELATIONSHIP] -> [ENTITY B]

Once the information is parsed, it can be stored in a graph database. The information stored is the knowledge base, not the original text. For information retrieval, the LLM needs to come up with an Entity query related to the question to retrieve the related entities and relationships. The retrieved information is much more concise and to the point than in the case of vector databases.

Tony Seale on his part argues that vectors need graphs. Embedding vectors are a pivotal tool when using Generative AI. While vectors might initially seem an unlikely partner to graphs, their relationship is more intricate than it first appears. And Ben Lorica makes the case for boosting LLMs with external knowledge using Knowledge Graphs.

Graph databases riding the RAG wave

RAG is an alternative to fine-tuning LLMs which seems less demanding and more immediately applicable. What’s more, it can be offered by vendors whose core competency is data management.

Since they already have a footprint in organizations, data management vendors can do the heavy lifting of integrating with LLMs. It’s a win-win: executives get to check their gen AI boxes, and vendors get to expand their offering, catching the buzz and keeping clients happy.

This is the reason why we’re seeing graph database vendors adding vector capabilities to their products. Neo4j was the first one to include vector capabilities in its offering in August 2023. The idea is to combine the best of both worlds as part of the roadmap newly appointed Neo4j CPO Sudhir Hasbe announced in July 2023.

Grounding LLMs with Neo4j

Neo4j followed that up by announcing a collaboration with AWS. Amazon Neptune followed suit, recently announcing vector search as part of its Neptune Analytics engine; see below.

Graph analytics meets HPC

Neo4j and Amazon Neptune seem to have been on similar trajectories. Another thing that both vendors unveiled in the last few months is new analytics engines, leveraging parallelism to achieve speedup in processing in graph analytics scenarios previously underserved.

As Hasbe shared, for Neo4j’s new analytics engine typical examples are analytical queries that traverse most of the graph data. The parallel runtime is specifically designed to address these analytical queries.

The new Neptune Analytics engine is geared towards three use cases. First, ephemeral analytics. These are workflows where customers just need to spin up a graph really quickly, run some analysis, and turn it off.

Second, low-latency analytical queries. That involves established machine learning pipelines with feature tables to perform real-time predictions. The third use case is building GenAI applications. Being able to perform vector similarity search when storing embeddings on Neptune analytics means it’s much easier to translate natural language questions into graph queries.

Amazon Neptune introduces a new Analytics engine and the One Graph vision

To power their new analytics engines, both teams seem to have borrowed a page from high performance compute (HPC). Neo4j’s implementation was directly inspired by the research paper “Morsel-Driven Parallelism: A NUMA-Aware Query Evaluation Framework for the Many-Core Age”.

While we don’t have specific pointers for Amazon Neptune’s implementation, Amazon Neptune General Manager Brad Bebee acknowledged certain similarities. Both teams see a broad subset of graph customers and graph use cases.

And both teams have members who are familiar with the literature and techniques coming from (HPC) processing of large scale graphs. The parallel processing and memory optimization techniques are things that are well-understood in the HPC community.

The difference is that HPC researchers are often solving a very specific graph problem on a very specific graph. For a service that has to solve graph problems for many customers, the main challenge is to generalize the techniques that can work well for HPC.

Graph database entries and exits

Even though Neo4j and Amazon are well-established in the graph database space, there is lots of action in this space beyond these two. Most vendors are working on positioning themselves in the generative AI landscape and/or adding conversational interfaces to their products.

But that does not mean that generative AI is all there is out there – there are other use cases as well. Aerospike Graph is a new entry in the graph database market, aiming to tackle complex problems at scale.

Aerospike started out as a key-value store. Eventually, the initial offering expanded to include the document model (JSON) as well as a SQL interface via Starburst. Graph was the next step, with adoption being reportedly customer-driven.

A team was built for Aerospike Graph, including Apache TinkerPop founder Marko Rodriguez and key contributors in the project. They helped Aerospike create a graph layer that interfaces with the core engine in a way that scales out horizontally in a shared-nothing architecture.

Aerospike Graph: A new entry in the graph database market, aiming to tackle complex problems at scale

Coincidentally, it was only a few days after Aerospike Graph’s official unveiling that another database vendor with a similar profile who had recently entered the graph database market announced its exit.

In 2019, Redis introduced their graph database, citing similar reasoning: they wanted to offer performance and scalability. In 2023, they wound down RedisGraph saying:

“Many analyst reports predicted that graph databases would grow exponentially. However, based on our experience, companies often need help to develop software based on graph databases. It requires a lot of new technical skills, such as graph data modeling, query composition, and query optimization. As with any technology, graph databases have their limitations and disadvantages.

This learning curve is steep. Proof-of-concepts can take much longer than predicted and the success rate can be low relative to other database models. For customers and their development teams, this often means frustration. For database vendors like Redis, this means that the total pre-sales (as well as post-sales) investment is very high relative to other database models”.

Subscribe to the Year of the Graph Newsletter

Keeping track of all things Graph Year over Year

Scoping and building knowledge graphs

Building knowledge graphs is supposedly a huge and terrifying project, as Mike Dillinger notes. But that perception is mostly coming from software engineers who think that a knowledge graph has to be enormous.

Tech organizations are overflowing with engineering managers who think (or expect) that engineers can do anything and everything. And there are precious few product managers who are familiar enough with knowledge graphs to frame knowledge graph building in terms that managers and engineers can grok and buy into.

The end result is that organizations don’t implement crucial but unfamiliar tech, like knowledge graphs. Contrary to widespread assumptions, Dillinger adds, creating and curating knowledge graphs is not an inherently manual process even if quality management requires expert review. He shares some of the latest research efforts for assisted knowledge graph creation.

Power Tools for Powerful Knowledge Graphs

The topic of assisted knowledge graph creation has received renewed attention in the last few months, again owing to the generative AI momentum. There are good reasons for this. LLMs can assist in converting any text into a graph of concepts, as Rahul Nayak goes to show.

One such effort that has drawn lots of attention is MechGPT, a language model specifically fine-tuned for constructing knowledge graphs. MechGPT first breaks down texts into small chunks. Each chunk is fed to a general-purpose LLM that generates question-answer pairs summarizing key concepts.

Bonus track 1: Yejin Choi in conversation with Bill Gates, summarized by Jesús Barrasa. Α combination of LLMs (magic but opaque / subsymbolic) with knowledge graphs (explicit / symbolic) is the path forward.

Bonus track 2: Prompting an LLM with an ontology to drive Knowledge Graph extraction from unstructured documents by Peter Lawrence.

Ontology modeling for Knowledge Graphs with SHACL

What is an ontology? That may seem like a trivial question for people familiar with knowledge graphs, but as Kurt Cagle goes to show, it is not. Cagle defines an ontology as a set of schemas that collectively establish the shape of the data held within a named graph. Others may prefer different definitions.

Holger Knublauch writes that in the world of knowledge graphs, an ontology is a domain model defining classes and properties. Classes are the types of entities (instances) in the graph and properties are the attributes and relationships between them. Ontologies define the structure of graphs and allow tools to make more sense of them.

Either way, there seems to be convergence around the notion of using SHACL, the SHApes Constraint Language, for Ontology Modeling. Prior to the introduction of SHACL, knowledge graph validation would mostly be manual, or reliant on OWL (Web Ontology Language) constraints. OWL constraints, however, are counter-intuitive.

Ontology Modeling with SHACL: Getting Started

Knublauch followed up on his SHACL tutorial with Part 2 on Qualified Cardinality Constraints. and Part 3 on SPARQL-based Constraints. Radostin Nanov has also written a SHACL guide in 3 parts: Learning how SHACL can tame unruly data graphs, applying SHACL to data and handling the output, and the internals of a SHACL engine. And Ivo Velitchkov with Veronika Heimsbakk maintain a Wiki on SHACL.

Bonus track 1: A Landscape of Ontologies Standards Landscape. This report presents a curated collection of ontologies that are highly relevant to ICT domains and vertical sectors, considering their maturity, prominence, and suitability for representing linked data in the semantic web. Ontologies can contribute in Explainable AI too. Bonus tracks 2 and 3: SPARQLing anything and a SPARQL Wiki.

Applied Graph AI use cases from DeepMind

DeepMind has been one of the forerunners of Graph AI. In the last few months, DeepMind has shared more details about their use of Graph AI in a couple of high-impact use cases.

In a paper published in Science, DeepMind introduces GraphCast, a state-of-the-art AI model able to make medium-range weather forecasts with unprecedented accuracy. GraphCast predicts weather conditions up to 10 days in advance more accurately and much faster than the industry gold-standard weather simulation system.

GraphCast is a weather forecasting system based on machine learning and Graph Neural Networks (GNNs), which are a particularly useful architecture for processing spatially structured data. DeepMind has open sourced the model code for GraphCast, enabling scientists and forecasters around the world to benefit billions of people in their everyday lives.

GraphCast: AI model for faster and more accurate global weather forecasting

DeepMind GNoME is a GNN-based system that discovered 2.2M new crystal structures including about 380k stable structures. Novel functional materials enable fundamental breakthroughs across technological applications from clean energy to information processing. Graph neural networks trained at scale can reach unprecedented levels of generalization, improving the efficiency of materials discovery by an order of magnitude.

Bonus track: Combinatorial optimization and reasoning with Graph Neural Networks. A conceptual review of recent key advancements in this emerging field, aiming at optimization and machine learning researchers by DeepMind’s Petar Veličković et.al.

Graph Levels of Detail

Graph-based data is ubiquitous in enterprise, across the industry verticals, and increasingly needed for machine learning use cases. Graph technologies are available, though not quite as widely used yet in comparison with relational databases. Even so, interest in knowledge graph practices has grown recently due to AI applications, given the benefits of leveraging graphs and language models together.

A frequent concern is that graph data gets represented at a low level, which tends to make queries more complicated and expensive. There are few mechanisms available — aside from visualizations — for understanding knowledge graphs at different levels of detail. That is to say, how can we work with graph data in more abstracted, aggregate perspectives?

Graph Levels of Detail

While we can run queries on graph data to compute aggregate measures, we don’t have programmatic means of “zooming out” to consider a large graph the way that one zooms out when using an online map. This leaves enterprise applications, which by definition must contend with the inherently multiscale nature of large scale systems, at a distinct disadvantage for leveraging AI applications.

Paco Nathan presents a survey of related methods to date for level-of-detail abstractions in graphs, along with indications toward future work.

Graph and Large Language Models research

Wrapping up with a collection of research on various aspects of combining graphs with large language models. Paco Nathan also lists a collection of research efforts to combine Graph ML with Language Models here.

IEEE researchers present a roadmap for unifying large language models and knowledge graphs. Their roadmap consists of three general frameworks. KG-enhanced LLMs, LLM-augmented KGs, and synergized LLMs + KGs. Cf. the previous YotG issue here.

Similarly, researchers from The Hong Kong University of Science and Technology (Guangzhou), The Chinese University of Hong Kong and Tsinghua University present a survey of graph meets large language model. They propose a taxonomy, which organizes existing methods into three categories based on the role (i.e., enhancer, predictor, and alignment component) played by LLMs in graph-related tasks.

Researchers from ETH Zurich, Cledar, and the Warsaw University of Technology introduce Graph of Thoughts (GoT): a framework that advances prompting capabilities in large language models (LLMs) beyond those offered by paradigms such as Chain-of Thought or Tree of Thoughts (ToT). Tony Seale summarizes the approach.

Michael Galkin et.al. present ULTRA, a single pre-trained reasoning model for Knowledge Graph reasoning. ULTRA can generalize to new KGs of arbitrary entity and relation vocabularies, which serves as a default solution for any KG reasoning problem. Researchers from Monash and Griffith present Reasoning on Graphs (RoG), a method that synergizes LLMs with KGs to enable faithful and interpretable reasoning.

A growing number of experts from across the industry, including academia, database companies, and industry analyst firms, like Gartner, point to Knowledge Graphs as a means for improving LLM response accuracy. To evaluate this claim, data.world researchers came up with a new benchmark that examines the positive effects that a Knowledge Graph can have on LLM response accuracy in the enterprise.

Generative AI Benchmark: Increasing the Accuracy of LLMs in the Enterprise with a Knowledge Graph

The benchmark compared LLM-generated answers to answers backed by a Knowledge Graph, via data stored in a SQL database. Findings point towards a significant improvement in the accuracy of responses when backed by a Knowledge Graph, in every tested category.

A*Net is a scalable, inductive and interpretable path-based Graph Neural Network on Knowledge Graphs. It can be used to make ChatGPT more factual by equipping it with knowledge graph reasoning tools. It’s open source and integrated with ChatGPT.

Google Research presents Talk like a Graph: Encoding Graphs for Large Language Models. A comprehensive study of encoding graph-structured data as text for consumption by LLMs.

Jure Leskovac et.al. introduce an end-to-end deep learning approach to directly learn on data spread across multiple tables called Relational Deep Learning. The core idea is to view relational tables as a heterogeneous graph, with a node for each row in each table, and edges specified by primary-foreign key relations.

The post Graphs, analytics, and Generative AI. The Year of the Graph Newsletter Vol. 25, Winter 2023 – 2024 appeared first on The Year of the Graph.

Return of the Graph: Geospatial Knowledge Graphs, Personal Knowledge Graphs, and Evolution. The Year of the Graph Newsletter...

George Anadiotis — Wed, 31 May 2023 09:01:15 +0000

New types of graphs, and a new era for the Year of the Graph Newsletter

The Year of the Graph Newsletter, keeping track of all things Graph year over year, is back after a long hiatus.

Read on to learn more about how the evolution of the newsletter follows the evolution of the domain and how to be involved, as well as industry news and analysis hot off the press:

The evolution of Graph and the Year of the Graph Newsletter
Knowledge Graphs are in conversational mode
Graph Database growth going strong through the Trough of Disillusionment
Graph AI is hot in research and making inroads into industry
Graph Analytics go big and realtime
Foursquare Graph, the first Geospatial Knowledge Graph
The Personal Knowledge Graph book, the first book on PKGs

The evolution of Graph and the Year of the Graph Newsletter

A lot of water has flowed under the bridge since the last issue of the YotG Newsletter. Two years is a long time in technology, especially for something as diverse and rapidly evolving as graphs. Here is a letter from the editor with a recap of what happened during the hiatus, and what’s coming next.

In terms of evolution of the Graph landscape, a few things have changed. The Newsletter may have taken a break, but keeping track of all things Graph Year over Year never stopped. The following sections assess where Knowledge Graphs, Graph Databases, Graph Analytics and Graph AI are today and where they’re headed to.

As you may know if you have been following the YotG Newsletter for a while, the editor here would be myself – George Anadiotis. I’ve been into graph in one way or another since the early 2000s. I’ve built Graph Database prototypes, was a part of award-winning Graph R&D, and led teams using Graph Databases in production.

In the last few years, i’ve been active as an analyst and writer covering Graph technology. I participated in the pivotal W3C Web Standardization for Graph Data initiative in Berlin as an independent expert and i have authored 3 reports and numerous articles on Graph Databases and beyond. I have also been organizing Connected Data World and compiling the YotG newsletter. And that’s not even all i do.

That’s a lot, so something’s got to give. This is what induced the hiatus, and why the YotG Newsletter is changing. The newsletter has historically been a long list featuring short excerpts from all-around industry news on all things Graph Analytics, Graph Databases, Knowledge Graphs, Graph AI and Data Science, compiled by myself exclusively. That was not sustainable.

The evolution of Graph and the Year of the Graph Newsletter

Going forward, the newsletter will be shorter, featuring longer excerpts from my own work on Graph, as well as a brief editorial and some analysis. Updates from industry and research are still published on the YotG Twitter and LinkedIn accounts. Feel free to follow those for the latest news, and reach out if you are interested in contributing.

The YotG web site has been revamped. It looks better and is easier to navigate, but the biggest change is that YotG’s list of resources on all things Graph Analytics, Graph Databases, Knowledge Graphs, and Graph AI are now available under the corresponding sections. Feel free to browse, search, learn, and share.

Fun fact #1: Classifying the huge backlog of 3K+ resources was done with a little help from ChatGPT.

Last but not least, you can always contact the YotG if you want to:

Ask a question or share an invite
Submit news items or otherwise contribute
Inquire about sponsored content or professional services

Knowledge Graphs are in conversational mode

The emergence of ChatGPT and the subsequent hype have made people across different domains think about ways they could use Large Language Model (LLM) technology to their benefit. Knowledge Graphs are no exception. As Dean Allemang puts it:

“Quite a lot [of what I’ve been reading about ChatGPT and LLMs] has to do with how the weaknesses of LLMs can be addressed by the use of [people’s] favorite technology. This is probably because I hang out in groups where people are interested in their own pet technology.

Since I have been an advocate for Knowledge Graph technology for many years, I am as guilty of this as the next writer, of thinking that the key to making LLMs useful is to combine them with Knowledge Graphs”.

There has been no shortage of ideas on how to combine Knowledge Graphs with LLMs. Most of them fall under one of three categories: using an LLM to create a new Knowledge Graph, using an LLM to access an existing Knowledge Graph, or using a Knowledge Graph to augment a LLM. There are numerous ongoing efforts and experiments, of which we are going to share just a few.

Harnessing the Power of Knowledge Graphs for Language Model Governance

GraphGPT creates a knowledge graph of the connections between people and proper nouns contained in input sentences. GraphGPT uses GPT-3 and was created by Varun Shenoy , a researcher at Stanford University.

Peter Lawrence shows how a knowledge graph can prompt or fine-tune a LLM enabling users to ask their questions. To illustrate this, he uses an RDF knowledge graph of a process plant, the core of a Digital-Twin, to prompt or fine-tune OpenAI’s GPT LLM. Fun fact #2: we called this back in 2020.

Tony Seale, Knowledge Graph Engineer at UBS, suggests harnessing the power of Knowledge Graphs for Language Model governance. This includes alignment with an ontological worldview, i.e. leveraging concepts defined in a Knowledge Graph’s ontology to constrain the language model.

Graph Database growth going strong through the Trough of Disillusionment

One of the first items in the YotG graph collection of graph resources was a Graph Database market forecast report by Markets and Markets. Back in early 2018, the report valued the Graph Database market size at $700 Million in 2017 and projected it to reach $2,4 Billion by 2023, at a Compound Annual Growth Rate (CAGR) of 24%.

That’s pretty substantial and pretty optimistic, but has it actually materialized?

In the most recent update of this report by Markets and Markets in 2022, the Graph Database market size is estimated at $1,9 Billion in 2021 and projected it to reach $5,1 Billion by 2026, at a CAGR of 22,5%.

Even though it’s not really possible to verify those estimates except against previous estimates, in that respect the projections seem pretty much on the mark. CAGR has declined somewhat but remains high and the market is still poised for growth. But it has not been uninterrupted growth all the way. If we take Gartner’s hype cycle at face value, Graph Databases are probably still going through the Trough of Disillusionment.

The Great Graph Debate: Revolutionary concept in databases or niche curiosity?

That means that the utility of graph databases may be questioned. In addition, recent and ongoing economic downturn has not left the graph database market unaffected. Some vendors were part of the wave of layoffs, others have changed leadership, funding has slowed down and the economic climate is reportedly having an impact on sales.

But that does not mean progress is stalling either. Investing in Graph Databases has not evaporated – far from it. And Graph Databases keep working on becoming more accessible, adding cloud services, visual interfaces and – you guessed it – conversational interfaces too.

RDF and LPG Graph Databases are converging through RDF-star, the RDF update that enables RDF to operate as property graphs, and GQL, the new international standard query language that is expected to be generally available in 2024.

If you are looking to decipher Graph Databases and make a decision on what is the right solution for you, check out the YotG Graph Database report for detailed vendor by vendor analysis.

Graph AI is hot in research and making inroads into industry

When the YotG first included Graph AI in its coverage, Graph AI was considered exotic. While Graph AI is still generating lots of research output, it’s also making inroads into industry.

Even back in 2019, Graph AI was a considerable part of new research in top AI venues. That trend has been accelerating; for example, check out this Guide to ICLR 2023, where Graph Representation Learning, Geometric Deep Learning and its applications for molecular modeling and physics are key themes.

Transformers, the architecture behind LLMs, can also be viewed as a Graph Neural Network. Graph Convolutional Networks (GCNs) combine deep learning with feature diffusion to produce useful node embeddings. And Geometric Deep Learning provides a common blueprint allowing to derive from first principles neural network architectures as diverse as CNNs, GNNs, and Transformers.

Neural Graph Databases: A new milestone in graph data management

Somewhere in between research and industry is the concept of Neural Graph Databases. Tailored for large incomplete graphs and on-the-fly inference of missing edges using graph representation learning, neural reasoning promises to maintain high expressiveness and support complex logical queries similar to standard graph query languages.

In its short history, Graph AI has had a strong footing in the pharma industry with applications in drug discovery. Recursion, a major player in drug discovery, has just acquired two startups in order to leverage their Graph AI-powered intellectual property: Valence Discovery (Mila, Montreal) for $47.5M and Cyclica (Toronto) for $40M. Beyond pharma, Graph AI is also used at the likes of Airbnb.

Graph Analytics go big and realtime

Graph Analytics is coming on its own. There are a few noteworthy pieces of evidence to back up this claim. First, there are now multiple market forecast reports that consider Graph Analytics a market of its own. For example, Research and Markets states that:

“The global graph analytics market in 2022 was valued at US$1.14 billion. The market value is anticipated to grow to US$6.90 billion by 2028. The market value is expected to grow at a CAGR of 34.80% during the forecast period of 2023-2028”.

Second, there are more and more resources around graph analytics, and they generate a lot of interest. Some recent examples: Maryam Miradi’s collection of 25 top Python libraries, types, algorithms and techniques for Graph Analytics. Amy Hodler’s Graph Analytics talk for G-Research in London.

Databricks faces critical strategic decisions. Here’s why.

More than anything else, however, it’s the fact that Graph Analytics are having an impact in the real world. From ESG to Customer 360, finance, supply chains, retail and anti-fraud, Graph Analytics is gaining adoption and becoming a significant part of vendor revenue. The importance of Graph Analytics may make it a central part of Databricks’ strategic decisions going forward too.

In industry use cases, the requirements often dictate processing large volumes of data in near-realtime. Graph analytics are growing in that direction, as exemplified by the likes of PayPal and LinkedIn. There are now solutions for streaming Graph Analytics, too. Plus, Graph Analytics are powering cybersecurity unicorns. We’re watching this space and we’ll be back with more details.

Foursquare moves to the future with a Geospatial Knowledge Graph

If the name Foursquare rings a bell, it means you were around in the 2010s. Your only resort to plausible deniability would be if you are a data professional – although that’s not an either/or proposition.

In the 2010s, Foursquare was a consumer-oriented mobile application. The premise was simple: people would check in at different locations and get gamified rewards. Their location data would be shared with Foursquare and used for services such as recommendations.

Facebook and Yelp got the lion’s share of that market, but Foursquare is still around. In addition to having 9 billion-plus visits monthly from 500 million unique devices, Foursquare’s data is used to power the likes of Apple, Uber and Coca-Cola.

Recently the company announced Foursquare Graph, what it dubs the industry’s first application of graph technology to geospatial data.

Foursquare moves to the future with a Geospatial Knowledge Graph

Heroes journey: Notes towards a Personal Knowledge Graphs Book

The last couple of years have seen the emergence of a new type of Knowledge Graphs: Personal Knowledge Graphs. Knowledge Graph definitions abound. Personal Knowledge Graph definitions are in flux.

Previously, Personal Knowledge Graphs were academically defined as graphs containing facts about a person, held by a 3rd party. That has recently changed,. More importantly, however, Personal Knowledge Graphs power a booming ecosystem of real-world tools focused on end-users, with data sovereignty and note-taking as key elements.

Note-taking is a timeless practice. Over time, a multitude of software tools have been developed to assist with note-taking. Applying Graph metaphors and principles in the personal information management domain and note-taking results in what we call Personal Knowledge Graphs.

Heroes journey: Notes towards a Personal Knowledge Graphs Book

The first book on Personal Knowledge Graphs is a journey of exploration and mapping of emergent practices and tools.

The hero’s journey is a common template of stories that involve a hero who goes on an adventure, is victorious in a decisive crisis, and comes home transformed. Writing a book is like a journey too. Writing the first Personal Knowledge Graphs book involved more than one hero and a few crises.

Would you like to receive the latest Year of the Graph Newsletter in your inbox? Easy – just signup below. Have some news you think should be featured in an upcoming newsletter? Easy too – reach out here.

The post Return of the Graph: Geospatial Knowledge Graphs, Personal Knowledge Graphs, and Evolution. The Year of the Graph Newsletter Vol. 24, Spring 2023 appeared first on The Year of the Graph.

DEV Community: George Anadiotis

Beyond Context Graphs: How Ontology, Semantics, and Knowledge Graphs Define Context. The Year of the Graph Newsletter Vol 30

Table of Contents

An Introduction to Context Graphs

Context Beyond Context Graphs

Ontologies, Context Graphs, and Semantic Layers: What AI Actually Needs in 2026

Tooling and Evaluation Frameworks for Ontologies

From Retrieval Augmented Generation to Knowledge Augmented Generation

Knowledge Graphs in Software Engineering and Enterprise Architecture

Knowledge Graph Research, Applications and Best Practices

Knowledge Graph Tools and Platforms

The State of the Graph Database Market

Graph Analytics and Graph AI Updates

The Ontology issue: From knowledge to graphs and back again. The Year of the Graph Newsletter Vol. 29, Winter 2025 – 2026

Table of Contents

The “O” word

From knowledge to graphs and back again

Knowledge graph applications at scale

Ontology and knowledge graph insights, tools and education

Two meanings of “Semantic Layer” and why both matter for AI

Subscribe to the Year of the Graph Newsletter

Graph databases: growing market, competition & options

New tools and research

Graph is the new star schema. The Year of the Graph Newsletter Vol. 28, Autumn 2025

Table of Contents

Graph is the new star schema

Graphs power Systems of Intelligence

A unified semantic knowledge graph for Enterprise AGI

Defining and building ontologies

Getting started with knowledge graphs

Adopting, building and populating knowledge graphs

Subscribe to the Year of the Graph Newsletter

Knowledge graphs and AI: a two-way street

Agentic knowledge graph construction and temporal graphs

The state of GraphRAG

Multimodal graphs

Graph databases grow and evolve

Graph data models: LPG vs. RDF, OWL vs. SHACL

Graph AI: GNNs, graph transformers and foundational models

Graph science: Strong perfect graphs, the new Dijkstra’s algorithm and convergent neural networks

The evolution of the graph technology and business landscape in 2025. Year of the Graph Newsletter V. 27, Spring–Summer 2025

Table of Contents

The graph technology landscape

Knowledge graph market outlook

Building and evaluating knowledge graphs as durable assets

Knowledge graphs powering household names through mergers and acquisitions

Subscribe to the Year of the Graph Newsletter

Knowledge graphs as the essential truth layer for Pragmatic AI

Graph RAG galore

New graph database engines, standardization and performance

Graph analytics and visualization: roadmap, features and platforms

Graph Foundation Models, applications at scale, LLMs and graphs

Graph Database market update, September 2024: Google Cloud Spanner Graph, Amazon Neptune, Neo4j

Google Cloud Spanner goes Graph

How does Spanner Graph work?

Querying RDF knowledge graphs with openCypher on Amazon Neptune

Neo4j gets Aura GenAI, self-service, performance and security features

The Graph Database market is moving forward

Knowledge Graph Enlightenment, AI and RAG. The Year of the Graph Newsletter Vol. 26, Summer 2024

Knowledge Graphs as Critical Enablers reaching the Slope of Enlightenment

Graph RAG: Approaches and Evaluation

Connected Data London: Bringing together Leaders and Innovators

Advances in Graph AI and GNN Libraries

Graph Database Market Growth and the GQL standard

Knowledge Graph Research, Use Cases and Data Models

Graphs, analytics, and Generative AI. The Year of the Graph Newsletter Vol. 25, Winter 2023 – 2024

Vectors, graphs, and RAG

Graph databases riding the RAG wave

Graph analytics meets HPC

Graph database entries and exits

Subscribe to the Year of the Graph Newsletter

Scoping and building knowledge graphs

Ontology modeling for Knowledge Graphs with SHACL

Applied Graph AI use cases from DeepMind

Graph Levels of Detail

Graph and Large Language Models research

Return of the Graph: Geospatial Knowledge Graphs, Personal Knowledge Graphs, and Evolution. The Year of the Graph Newsletter...

The evolution of Graph and the Year of the Graph Newsletter

Knowledge Graphs are in conversational mode

Graph Database growth going strong through the Trough of Disillusionment