How lawliet119/kg-movie-recommender works
KGenSam is a knowledge-graph conversational movie recommender, with a browser chat/visualization frontend and a FastAPI recommendation backend. The main workflow is user chat → backend session and recommendation handling → preference-aware ranking → movie recommendations and feedback. The repository also includes interaction-data/model training and offline evaluation capabilities. The graph reflects explicit imports and documented frontend/backend responsibilities; some internal endpoint wiring is not shown in the excerpts, so uncertain links are omitted.
Main entry points: main.py FastAPI endpoints · recommender.py Recommendation engine · chat.js Chat interface · interaction_data.py Interaction profiles
Components 25
Browser experience
- Chat interface
src/ui/chat.js - App entry
src/main.js - Graph visualization
src/ui/graphViz.js - Browser movie graph
src/data/movieKG.js - Browser recommender
src/engine/recommender.js
API and conversation
- FastAPI endpoints
backend/app/main.py - Conversation sessions
backend/app/conversation_manager.py
Knowledge and models
- Semantic NLU
backend/app/semantic_nlu.py - Movie knowledge graph
backend/app/kg_builder.py - MovieLens and TMDB KG
backend/app/tmdb_kg.py - KG embeddings
backend/app/kg_embeddings.py - Interaction profiles
backend/app/interaction_data.py
Recommendation
- Recommendation engine
backend/app/recommender.py - Active question sampler
backend/app/active_sampler.py - Ask-recommend policy
backend/app/interact_policy.py - Preference propagation
backend/app/preference_propagator.py - Factorization model
backend/app/fm_model.py - Fusion ranker
backend/app/fusion_ranker.py - Negative sampler
backend/app/negative_sampler.py
Training and evaluation
- Policy training
backend/app/policy_training.py - Offline evaluation
backend/app/evaluation.py - User simulator
backend/app/user_simulator.py - Runtime checkpoints
backend/app/model_checkpoint.py
Other components
- Movie seeker
- Gemini Graph-RAG
Connections 30
- Movie seeker → Chat interface: sends chat
src/ui/chat.js - App entry → Chat interface: initializes
src/main.js - App entry → Browser movie graph: builds graph
src/main.js - App entry → Browser recommender: creates engine
src/main.js - Chat interface → Graph visualization: updates graph
src/ui/chat.js - Chat interface → Browser recommender: local fallback
src/ui/chat.js - Chat interface → FastAPI endpoints: requests backend
- FastAPI endpoints → Conversation sessions: manages sessions
backend/app/main.py - FastAPI endpoints → Recommendation engine: serves recommendations
backend/app/main.py - FastAPI endpoints → Movie knowledge graph: loads graph
backend/app/main.py - FastAPI endpoints → KG embeddings: loads embeddings
backend/app/main.py - FastAPI endpoints → Semantic NLU: loads language model
backend/app/main.py - FastAPI endpoints → Interaction profiles: loads interactions
backend/app/main.py - MovieLens and TMDB KG → Movie knowledge graph: builds graph data
- Recommendation engine → Movie knowledge graph: traverses graph
backend/app/recommender.py - Recommendation engine → KG embeddings: scores embeddings
backend/app/recommender.py - Recommendation engine → Active question sampler: selects questions
backend/app/recommender.py - Recommendation engine → Ask-recommend policy: chooses action
backend/app/recommender.py - Recommendation engine → Preference propagation: propagates preferences
backend/app/recommender.py - Recommendation engine → Factorization model: scores candidates
backend/app/recommender.py - Recommendation engine → Fusion ranker: ranks features
backend/app/recommender.py - Negative sampler → Interaction profiles: uses profiles
backend/app/negative_sampler.py - Factorization model → Interaction profiles: trains from feedback
backend/app/fm_model.py - Policy training → Ask-recommend policy: trains policy
backend/app/policy_training.py - Policy training → Interaction profiles: uses profiles
backend/app/policy_training.py - Offline evaluation → Recommendation engine: evaluates ranking
backend/app/evaluation.py - Offline evaluation → User simulator: simulates users
backend/app/evaluation.py - FastAPI endpoints → Offline evaluation: runs evaluation
backend/app/main.py - FastAPI endpoints → Runtime checkpoints: loads checkpoint
backend/app/main.py - Chat interface → Gemini Graph-RAG: requests grounded answer
src/ui/chat.js