Apppublic

sampstad/narrative_detector

sourceHugging Faceupdated 6mo agoView on Hugging Face

0likes

App README

📰 Narrative Extractor

Extract the key narrative elements from any news article or paragraph — who, what, when, where, and why — with semantic thread clustering to separate distinct storylines.

What it extracts

Section	Method
Summary	Extractive (top-2 entity-rich sentences)
Tone	DistilBERT (`distilbert-base-uncased-finetuned-sst-2-english`)
People / Orgs / Places	spaCy NER (`en_core_web_lg`) + FastCoref coreference
Time references	spaCy DATE/TIME entities + dateparser normalisation
Key actions	Subject-Verb-Object triples from dependency parse
Causal links	Regex on connectives with confidence scores
Narrative threads	`all-MiniLM-L6-v2` embeddings + HDBSCAN clustering

Run locally

bash

pip install -r requirements.txt
python app.py

Project structure

narrative_HF/
├── app.py            # Streamlit UI
├── pipeline.py       # NLP extraction logic
├── requirements.txt  # Python dependencies
└── .streamlit/
    └── config.toml   # Theme configuration