Broad intuition · Study map 20 of 85
Graphs and recommenders: learning from relationships
Suggested preparation: Audio, video, and multimodal models
Picture it
A graph model updates an item using information from its neighbors. A recommender often retrieves a few hundred candidates cheaply, then spends more computation ranking them. Both rely on relationships as much as individual features.
What to understand
- Message passing and graph neural networks
- Candidate retrieval, ranking, feedback loops, and cold start
Follow the references
1 · Course
Stanford · CS224W machine learning with graphs ↗Start with graph representations and message-passing lectures.
2 · Course
Google · Recommendation systems ↗Follow candidate generation → scoring → re-ranking.
You understand it when…
Why should a recommender avoid running its most expensive ranker over every item?
This is a study map. Follow the reference for the full explanation; use the question to test your mental model. Browse the reference shelf.