How dracula-5/news-recommender works
NeuroFeed is an adaptive news-recommendation product: users browse a feed, and each interaction can update their profile and influence subsequent recommendations. The principal path runs from the browser through the FastAPI service to the recommender, which retrieves and scores candidates, blends cold-start and learning signals, diversifies results, and persists interaction state in SQLite. Separate branches cover authentication, live-news ingestion, webcam attention, and evaluation. Connections involving unsampled modules are limited to relationships documented in the README or listed in the source index.
Main entry points: recommender.py Recommendation engine · main.py FastAPI service · database.py News and interaction store · live_news.py Live-news ingestion
Components 19
User experience
- Feed UI
final/frontend/app.js
API service
- FastAPI service
final/backend/main.py - Authentication
final/auth.py - TTL cache
final/cache.py - Runtime config
final/config.py
Recommendation
- Recommendation engine
final/recommender.py - Ensemble retrieval
final/nlp_engine.py - MMR reranking
final/reranking.py - Mood signals
final/mood_signals.py - Interest graph
final/graph_memory.py - Webcam attention
final/webcam_attention.py
State and ingestion
- Live-news ingestion
final/live_news.py - News and interaction store
final/database.py - Auth token store
final/auth.py
Learning and evaluation
- Dueling DDQN
final/ddqn_agent.py - Hybrid bandits
final/rl_policies.py - Offline evaluation
final/evaluation.py
Other components
- Browser user
- NewsAPI
Connections 25
- Browser user → Feed UI: uses
README.md - Feed UI → FastAPI service: REST requests
README.md - FastAPI service → Recommendation engine: routes requests
final/backend/main.py - FastAPI service → Authentication: authenticates users
final/backend/main.py - FastAPI service → TTL cache: uses cache
final/backend/main.py - FastAPI service → Live-news ingestion: routes live news
final/backend/main.py - FastAPI service → Offline evaluation: routes evaluation
final/backend/main.py - FastAPI service → Webcam attention: reads attention
final/backend/main.py - FastAPI service → Runtime config: loads settings
final/backend/main.py - Recommendation engine → Ensemble retrieval: scores candidates
final/recommender.py - Recommendation engine → MMR reranking: diversifies results
final/recommender.py - Recommendation engine → Mood signals: uses mood signals
final/recommender.py - Recommendation engine → Interest graph: uses interest graph
final/recommender.py - Recommendation engine → Dueling DDQN: uses Q-values
final/recommender.py - Recommendation engine → Hybrid bandits: uses bandit signals
final/recommender.py - Recommendation engine → Webcam attention: reads attention
final/recommender.py - Recommendation engine → News and interaction store: reads and writes
final/recommender.py - Interest graph → News and interaction store: persists graph state
final/graph_memory.py - Dueling DDQN → Hybrid bandits: blends policy
final/ddqn_agent.py - Dueling DDQN → News and interaction store: loads agent state
final/ddqn_agent.py - Offline evaluation → Recommendation engine: evaluates rankings
final/evaluation.py - Offline evaluation → News and interaction store: reads catalog
final/evaluation.py - Live-news ingestion → NewsAPI: fetches headlines
final/live_news.py - Live-news ingestion → News and interaction store: upserts articles
final/live_news.py - Authentication → Auth token store: stores credentials
final/auth.py