Story Clustering

This code takes newsitems in the format as provided by Taranis AI and clusters them into Stories.

Description and Use

The approach supports the following functionalities:

1. Automatically detect Events.
2. News items are clustered based on the detected Events.
3. Documents belonging to related Events are then clustered into Stories.

Initial clustering

The method initial_clustering in clustering.py takes as input a dictionary of news_items_aggregate (see tests/testdapa.py for the actual input format) and outputs a dictionary containing two keys: ("event_clusters" : list of list of documents ids) and ("story_clusters" : list of list of documents ids)

Incremental clustering

The method incremental_clustering_v2 takes as input a dictionary of news_items_aggregate, containing new news items to be clustered, and clustered_news_items_aggregate, containing already clustered items, and tries to cluster the new documents to the existing clusters or create new ones. See tests/testdata.py for the actual input formats. This method also outputs a dictionary containing two keys: ("event_clusters" : list of list of documents ids) and ("story_clusters" : list of list of documents ids)

Usage

flask run
# or
granian run
# or
docker run -p 5000:5000 ghcr.io/taranis-ai/taranis-story-cluster-bot:latest

Development

uv venv
uv sync --all-extras --dev

See notebook\test_story_clustering.ipynb for examples on how to use the clustering methods.

License

EUROPEAN UNION PUBLIC LICENCE v. 1.2