How to Detect Investing Sentiment Anomalies with the Pulsebit API (Python)

How to Detect Investing Sentiment Anomalies with the Pulsebit API (Python)

We recently discovered an intriguing anomaly in our sentiment data: a 24-hour momentum spike of -0.248 concerning the investing topic. This significant negative shift indicates a sudden drop in positive sentiment, particularly highlighted by the absence of articles covering the related news. The cluster story that emerged reads, "BofA raises South Africa rate outlook on oil-driven inflation risk - Investing.c." This tells us there might be a deeper narrative at work, especially with the clear focus on South Africa's economic outlook.

The Problem

The challenge here lies in the structural gap many sentiment analysis pipelines face when they don't account for multilingual origins or entity dominance. Your model may have missed this critical insight by hours, potentially overlooking the urgency of the South African economic situation. With the dominant language being English in this context, we can't afford to ignore the implications of such a gap in our data processing.

en coverage led by 19.8 hours. id at T+19.8h. Confidence scores: en 0.86, es 0.85, fr 0.85 Source: Pulsebit /sentiment_by_lang.

The Code

Here's how we can catch this anomaly using our API in Python:

import requests

![Left: Python GET /news_semantic call for 'investing'. Right:](https://pub-c3309ec893c24fb9ae292f229e1688a6.r2.dev/figures/g3_code_output_split_1773271099911.png)
*Left: Python GET /news_semantic call for 'investing'. Right: returned JSON response structure (clusters: 3). Source: Pulsebit /news_semantic.*


# Define the parameters for the query
topic = 'investing'
score = +0.000
confidence = 0.00
momentum = -0.248

# Geographic origin filter (if available)
geo_filter = {
    'region': 'south_africa',  # You can specify the region here
}

# Fetch the sentiment data
response = requests.post("https://api.pulsebit.com/sentiment", json={
    "topic": topic,
    "score": score,
    "confidence": confidence,
    "momentum": momentum,
    "geo_filter": geo_filter
})

# Check the response
if response.status_code == 200:
    sentiment_data = response.json()
    print(sentiment_data)
else:
    print("Error fetching sentiment data:", response.text)

# Meta-sentiment moment
cluster_reason = "Clustered by shared themes: raises, south, africa, rate, outlook."
meta_response = requests.post("https://api.pulsebit.com/sentiment", json={"text": cluster_reason})

if meta_response.status_code == 200:
    meta_sentiment_data = meta_response.json()
    print(meta_sentiment_data)
else:
    print("Error fetching meta sentiment data:", meta_response.text)

In the above code, we first set our parameters for the investing topic and the negative momentum. The geographic filter allows us to zero in on South Africa, which is particularly useful when more data about language and country is available. If our dataset includes geo-filter data, we can enhance our analysis significantly.

Geographic detection output for investing. hk leads with 1 articles and sentiment +0.85. Source: Pulsebit /news_recent geographic fields.

Next, we take the cluster reason string and run it back through our API to determine how it frames the narrative itself. This additional layer allows us to assess if the sentiment aligns with the emerging themes.

Three Builds Tonight

With this anomaly in mind, here are three specific builds we can implement based on our findings:

1. Geographically Filtered Alerts: Create an alert system that triggers when momentum drops below a certain threshold (like -0.2) specifically for South Africa. This enables you to react quickly to economic news that could impact sentiment.

2. Meta-Sentiment Analysis Dashboard: Build a dashboard that visualizes the results of meta-sentiment analysis for various clusters. This can help you understand how narratives are shifting over time based on key themes, like those surrounding South Africa's economic outlook.

3. Cross-Language Sentiment Comparison: Develop a comparative analysis tool that evaluates sentiment across different languages for the same topic. This could uncover unique insights, particularly in diverse regions where multiple languages are spoken, enhancing your understanding of the sentiment landscape.

Get Started

If you're ready to dig into sentiment anomalies and build actionable insights, check out our documentation at pulsebit.lojenterprise.com/docs. You can copy-paste the code shared here and run it in under 10 minutes. Let's harness this data thoughtfully!