EVENT-CONTEXT is an automated news analysis system that collects global headlines from RSS feeds, applies AI-driven topic modeling (BERTopic), and generates daily intelligence reports with interactive visualizations.
Designed for researchers, analysts, and globally-minded citizens, it transforms raw news into structured insights — revealing trends, topic relationships, and global event dynamics.
🔍 URL: github.io/event-context/
- ✅ Daily news collection from international RSS sources
- 🔍 Topic modeling using BERTopic (NLP + sentence transformers)
- 📊 Interactive visualizations: topic evolution, co-occurrence, semantic similarity
- 📅 Historical tracking with date-based reports
- 🖼️ Self-contained HTML reports with embedded plots
- 🤖 Fully automated via GitHub Actions (runs every 24h)
- 🌐 GitHub Pages ready: no external hosting required
- 💾 Open & transparent: all data and code are versioned
flowchart TD
A[📡 1. Load RSS Feeds] --> B[🌐 2. Fetch News]
B --> C[🤖 3. BERTopic Modeling]
C --> D[📊 4. Topic Analysis]
D --> E[🔗 Co-occurrence Matrix]
D --> F[🧠 Semantic Similarity]
D --> G[📈 Topic Evolution]
E --> H[📂 Save Results]
F --> H
G --> H
H --> I[🌍 Generate HTML Report]
I --> J[📝 Update Index Page]
J --> K[🚀 GitHub Pages Deployment]
Each day, you get:
- A topic model with labels like
"China-Taiwan Tensions"or"Global Energy Crisis" - A time series chart showing how topics evolve
- A co-occurrence matrix of related topics
- A semantic similarity heatmap
- A daily HTML report with all insights
- An index page linking to historical reports
| Component | Technology |
|---|---|
| News Collection | feedparser + RSS |
| Topic Modeling | BERTopic + sentence-transformers |
| NLP | all-MiniLM-L6-v2 (embedding model) |
| Visualization | plotly, seaborn, matplotlib |
| Automation | GitHub Actions |
| Output Format | Static HTML (GitHub Pages compatible) |
| Hosting | GitHub Pages (docs/ folder) |
- Multilingual news processing – Extend analysis to non-English RSS feeds using multilingual embedding models (e.g.,
paraphrase-multilingual-MiniLM-L12-v2) - GDELT integration – Incorporate structured global event data from the GDELT Project for deeper geopolitical context
- Sentiment analysis per topic – Add emotional tone detection using VADER or transformer-based sentiment models
- Topic alert system – Get notified when specific topics (e.g., "conflict", "elections") emerge or spike
- Searchable archive – Add full-text search across historical reports using Lunr.js or FlexSearch
- Geospatial visualization – Map news coverage by country using interactive maps (Plotly, Leaflet)
- User feedback loop – Allow readers to label topics or relevance for semi-supervised improvement
JLVM
GitHub @jluisvim