Neural Networks for Recommender Systems
summary
Claims that neural nets reproduce and extend classic methods: matrix factorization = dot product of embeddings and replacing the dot with an MLP yields more flexible interactions. Details include using categorical embeddings trained end-to-end, triplet loss / Siamese networks for implicit feedback, and a YouTube-style session embedding that aggregates past item embeddings plus metadata. Mentions practical tooling (optimizers, dropout, pre-trained convnets) and Spotify/YouTube examples rather than abstract promises.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:19Neural Networks for Recommender Systems
- 6:53Matrix Factorization for CF
- 8:32Symbolic variables Recommender Systems
- 9:47One-hot representation
- 14:29Deep RecSys Architecture
- 15:09Deep RecSys with metadata
- 15:56Implicit Feedback: Triplet loss
- 17:19Deep Triplet Networks
- 19:10Embed all the things!
related talks
33mNeural Networks for Recommender Systems
29mFurther enhancements of recommender systems using deep learning
32mLet's talk about Recommendation Systems!
30mDeep Learning for Recommender Systems
20mBuilding Recommender system
22mNetflix's Big Bet: One model to rule recommendations
42mDeep Learning Recommender Architecture: Master Thesis to ...
18mBuilding an Incrementally Trained, Local Taste Aware, Global Deep Learned Recommender System Model