TechTalk "Not by Time Spent Alone: How We Build a Ranking System"
summary
A talk describing how a recommendation and ranking system is built: it covers the system architecture, the need for post-ranking (re-ranking) even when the base model has good quality, and techniques to correct popularity bias toward certain authors. It discusses practical measures to implement re-ranking and mitigate model skew in a high-load production environment.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 0:21Team and tasks
- 1:04Recommendation system structure
- 2:08Response time limits
- 3:12Popularity problem cycle
- 4:16Business impact on authors
- 5:18Reranking frequency
- 6:00Data skew solution
- 7:02TikTok comparison
- 8:28New author support
- 9:11Fraud prevention
- 9:52Resource allocation
- 10:34Conclusion
related talks
41mDynamic Ranking of Search and Recommendation Results in High-Traffic Systems
33mPeople who liked this talk also liked ... Building Recommendation...
41mRecommender Systems: Crafting Effective Personalised Experiences
55mLearning to Rank, a Little
01h 14mBuilding a Recommendation System with a Hybrid Search From Scratch
38mLearning to Rank: where search meets machine learning #bbuzz
23mML-Powered Search & Recommendation 101
32mLet's talk about Recommendation Systems!