Deep Learning Recommender Architecture: Master Thesis to ...
summary
A deep learning recommender evolves from a master thesis into a production pipeline that embeds users and items as 77‑float vectors, trains three nets (user, item, rank) in TensorFlow, and serves predictions via a Dockerized TensorFlow Serving + Scala gateway combo. Items (2M) stream through Kafka, embeddings update in real time, an hourly job rebuilds a ~125GB approximate nearest‑neighbor index with Spotify’s annoy, and the system narrows to ~200 candidates then reranks with the rank net; TensorFlow Serving shows ~2–3ms 95th‑percentile per request.
AI-written summary · about
chapters (8)
- 0:00Intro
- 1:01Motivation
- 5:03Components - User Preferences
- 6:01Deep Learning Approach
- 9:35Deep Learning Architecture Python to Scala
- 16:34Deep Learning Architecture Candidate Items Selection
- 26:24Recommendations Generator
- 32:23Final Notes
related talks
30mDeep Learning for Recommender Systems
58mRecommender Systems – A Walkthrough
29mFurther enhancements of recommender systems using deep learning
18mBuilding an Incrementally Trained, Local Taste Aware, Global Deep Learned Recommender System Model
01h 21mUnlocking Personalization: A Deep Dive into Modern Recommendation Algorithms
57mHands-On with LLM-Powered Recommenders: Hybrid Architectures
41mRecommender Systems: Crafting Effective Personalised Experiences
33mNeural Networks for Recommender Systems