How sr-ntr/cold-start-recommender works
This repository is an early-stage, cold-start-aware e-commerce recommendation research project, not yet a serving recommendation system. Its implemented workflow inspects and preprocesses Amazon Electronics interactions, then builds text embeddings from product metadata; the README describes popularity, collaborative, content-based, and hybrid recommendation approaches plus warm/low-history/cold-start evaluation as planned work. Edges reflect the shown preprocessing and embedding calls, or explicit README statements. The data files and recommendation/evaluation implementations are not in the supplied tree.
Main entry points: generate_embeddings.py Embedding generation · README.md Recommendation approaches · prepare_interactions.py Interaction preparation · inspect_interactions.py Interaction inspection
Components 18
Interaction data
- Raw interactions
- Interaction inspection
src/data/inspect_interactions.py - Interaction preparation
src/data/prepare_interactions.py - Processed interactions
Product embeddings
- Product catalog
- Product metadata
- Metadata loading
src/features/generate_embeddings.py - Product text shaping
src/features/product_text.py - Embedding generation
src/features/generate_embeddings.py - Sentence transformer
- Product embeddings
Recommendation research
- Recommendation approaches
README.md - Popularity baseline
README.md - Collaborative filtering
README.md - Content-based recommendations
README.md - Hybrid recommendations
README.md - Cold-start evaluation
README.md
Other components
- Researcher
Connections 17
- Researcher → Interaction inspection: runs
src/data/inspect_interactions.py - Raw interactions → Interaction inspection: reads
src/data/inspect_interactions.py - Researcher → Interaction preparation: runs
src/data/prepare_interactions.py - Raw interactions → Interaction preparation: reads
src/data/prepare_interactions.py - Interaction preparation → Processed interactions: writes
src/data/prepare_interactions.py - Researcher → Embedding generation: runs
src/features/generate_embeddings.py - Embedding generation → Product catalog: loads
src/features/generate_embeddings.py - Embedding generation → Metadata loading: calls
src/features/generate_embeddings.py - Metadata loading → Product metadata: queries
src/features/generate_embeddings.py - Embedding generation → Product text shaping: calls
src/features/generate_embeddings.py - Embedding generation → Sentence transformer: encodes
src/features/generate_embeddings.py - Embedding generation → Product embeddings: writes
src/features/generate_embeddings.py - Recommendation approaches → Popularity baseline: includes
README.md - Recommendation approaches → Collaborative filtering: includes
README.md - Recommendation approaches → Content-based recommendations: includes
README.md - Recommendation approaches → Hybrid recommendations: includes
README.md - Recommendation approaches → Cold-start evaluation: compares on
README.md