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

We recently came across an intriguing anomaly: a 24h momentum spike of +1.650. It’s a significant shift that can indicate a sudden change in sentiment within the markets. This kind of spike can provide us with valuable insights, but it also highlights the importance of monitoring sentiment data closely to stay ahead of trends.

In our experience, many pipelines fall short when they don't account for multilingual origins or the dominance of specific entities. Your model might have missed this critical spike by several hours, potentially leading to missed trading opportunities or misread market signals. For instance, if the dominant language of incoming sentiment data is English and your model is only set to analyze data in that language, it might overlook significant events unfolding in other languages, especially in regions where sentiment shifts occur first.

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

Here’s how we can catch this momentum spike using our API with Python:

import requests

![Left: Python GET /news_semantic call for 'markets'. Right: r](https://pub-c3309ec893c24fb9ae292f229e1688a6.r2.dev/figures/g3_code_output_split_1773144416092.png)
*Left: Python GET /news_semantic call for 'markets'. Right: returned JSON response structure (clusters: 0). Source: Pulsebit /news_semantic.*


# Define the parameters
topic = 'markets'
score = +0.000
confidence = 0.87
momentum = +1.650

# Geographic origin filter (Example: filtering for English language)
url = f"https://your.api.endpoint/v1/data?topic={topic}&language=en"
response = requests.get(url)
data = response.json()

# Check if geographic filtering is applicable
if data['geo_filter'] is None:
    print("DATA UNAVAILABLE: no geo filter data returned — verify /dataset/daily_dataset and /news_recent for topic: markets")
else:
    print("Geo filter applied successfully.")

![[DATA UNAVAILABLE: countries — verify /news_recent is return](https://pub-c3309ec893c24fb9ae292f229e1688a6.r2.dev/figures/g3_geo_output_1773144416176.png)
*[DATA UNAVAILABLE: countries — verify /news_recent is returning country/region values for topic: markets]*


# Meta-sentiment moment: Analyze the narrative framing
meta_sentiment_url = "https://your.api.endpoint/v1/sentiment"
narrative_input = "Markets narrative sentiment cluster analysis"
narrative_response = requests.post(meta_sentiment_url, json={"input": narrative_input})
narrative_analysis = narrative_response.json()

print("Meta-sentiment analysis:", narrative_analysis)

In the code above, we set up a geographic filter to query sentiment data based on language or country. If that data isn't available, we receive a prompt to check our dataset, ensuring we stay informed about the limitations of our inputs. We also run a meta-sentiment analysis on the narrative framing itself, providing insights into how the sentiment is structured. This is crucial as it allows us to grasp not just the data, but the context behind it.

Now, let’s look at three specific builds we can implement with this pattern:

1. Spike Detection with Geo-Filter: Set up an alert system that triggers when a 24h momentum spike exceeds a threshold (e.g., +1.500) for a specific language region (e.g., Spanish). This allows you to catch localized spikes before they spread.

2. Dynamic Sentiment Adjustment: Create a function that recalibrates your sentiment scoring based on the output of the meta-sentiment analysis. If your analysis suggests a low confidence score in the narrative, adjust your sentiment thresholds accordingly to avoid false positives.

3. Sentiment Heatmap Visualization: Combine the geographic data with momentum spikes to build a heatmap that visualizes sentiment shifts across different regions. Set a threshold for momentum (e.g., +1.650) to highlight areas with significant sentiment changes.

If you're ready to dive into sentiment data, check out our documentation at pulsebit.lojenterprise.com/docs. You can copy and paste the code snippets above and have them running in under 10 minutes. This is how we stay ahead of the curve in a data-driven landscape.