Unleashing the Power of Data Science
summary
Modern recommender systems scale by turning user–item interactions into graphs and training GNNs that learn embeddings; Pinterest’s pinsage runs on 17.5 billion examples, 3 billion nodes and 18 billion edges, serves millions daily, and uses six practical techniques: neighbor sampling, producer–consumer minibatches, distributed map‑reduce serving, biased random walks, importance pooling (attention), and curriculum training to handle multimodal signals, reduce memory, prioritize relevant neighbors, and meet strict latency and relevance constraints.
AI-written summary · about
chapters (13)
- 0:00Introduction and agenda
- 1:02What are recommender systems
- 2:46Traditional algorithms overview
- 4:32Limitations of traditional methods
- 5:59Introducing graph neural networks
- 8:10How graphs solve cold start problem
- 8:55Graph neural networks behind the scenes
- 10:21Case study introduction
- 10:44Pinsage algorithm explained
- 13:33Six techniques in Pinsage
- 17:09Real-world lessons from Pinterest
- 18:57Limitations of graph neural networks
- 20:00Q&A and closing
related talks
58mRecommender Systems – A Walkthrough
28mThe Power of Personalized Customer Experiences Through AI Innovations
30mDeep Learning for Recommender Systems
01h 21mUnlocking Personalization: A Deep Dive into Modern Recommendation Algorithms
33mA Sommelier's Guide to Recommendation Algorithms: Classical and Graph-Based Recommender Systems
41mRecommender Systems: Crafting Effective Personalised Experiences
42mDeep Learning Recommender Architecture: Master Thesis to ...
32mLet's talk about Recommendation Systems!