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

We’ve just uncovered a striking data anomaly: a 24-hour momentum spike of +1.000 in business sentiment. This spike indicates a significant shift in sentiment that should not be ignored. When you see a value this high, it triggers our curiosity—what's driving this sudden change? Understanding the underlying reasons demands a deeper look into the data.

When we examine our sentiment pipeline, a glaring gap stands out—multilingual origin and entity dominance are often overlooked. Your model probably missed this spike by several hours, failing to capture the nuances in sentiment across languages or dominant entities. For instance, if the primary language of the sentiment trend shifts from English to Spanish, or if a key entity suddenly emerges in a dominant position, your analysis could miss critical insights. This is especially relevant if the leading language is English and the spike occurs in a non-English context, indicating that your pipeline isn't structured for multilingual adaptability.

![DATA UNAVAILABLE: lag_hours — verify /dataset/daily_dataset
[DATA UNAVAILABLE: lag_hours — verify /dataset/daily_dataset is returning sentiment_by_lang data for topic: business]

To catch this anomaly, we can leverage our API efficiently. Below is a Python snippet that illustrates how to filter sentiment data based on geographic origin—provided that language and country data is available. In this case, we're focusing on the business topic with specific sentiment values:

![DATA UNAVAILABLE: countries — verify /news_recent is return
[DATA UNAVAILABLE: countries — verify /news_recent is returning country/region values for topic: business]

import requests

![Left: Python GET /news_semantic call for 'business'. Right: ](https://pub-c3309ec893c24fb9ae292f229e1688a6.r2.dev/figures/g3_code_output_split_1773141452130.png)
*Left: Python GET /news_semantic call for 'business'. Right: returned JSON response structure (clusters: 0). Source: Pulsebit /news_semantic.*


# Define the parameters for the query
topic = 'business'
score = +0.000
confidence = 0.87
momentum = +1.000

# Geographic origin filter (assuming we have geo data)
geo_filter = {'region': 'us'}  # Example filter for the US
response = requests.get(f'https://api.example.com/sentiment?topic={topic}&geo={geo_filter}')
data = response.json()

# Check the returned sentiment data
print(data)

In the above code, we set up a call to our sentiment endpoint, filtering by region. Remember, geo filtering is only possible when you have access to language or country data. If you don't, you'll need to ensure your dataset includes these details—check the /dataset/daily_dataset and /news_recent endpoints.

Next, to add depth to our analysis, we can run a meta-sentiment moment. This involves scoring the narrative framing itself by submitting the cluster reason string back through our sentiment scoring endpoint. Here’s how you can do it:

# Define the narrative framing
narrative = "Business narrative sentiment cluster analysis"
response_meta = requests.post('https://api.example.com/sentiment', json={'text': narrative})
meta_data = response_meta.json()

# Output the scored narrative
print(meta_data)

By analyzing the narrative itself, we can glean insights about the framing of the sentiment data. This adds a layer of context that pure numerical analysis alone can't provide.

Now that we have a method to detect anomalies, here are three specific builds we can implement:

1. Geographic Anomaly Detection: Set a threshold for momentum spikes over +0.500 for a specific region, triggering alerts when such spikes occur. This will help you catch significant shifts in sentiment before they impact your business strategy.

2. Meta-Sentiment Framing Analysis: Use the meta-sentiment loop to analyze the framing of narratives every time a momentum spike exceeds +0.750. This helps identify not only the sentiment but also how it’s being discussed in the media.

3. Cross-Language Monitoring: Implement a filter that alerts you when sentiment spikes in non-English regions with a minimum confidence score of 0.85. This will ensure you are aware of emerging global trends that could affect your business.

To get started with these insights and tools, check out our documentation. You can copy-paste the code snippets above and run them in under 10 minutes. Let’s make sure we don’t miss the next big sentiment shift!