Dynamic Ranking of Search and Recommendation Results in High-Traffic Systems
summary
Describes a system for dynamic ranking of search and recommendation results in high-load job platforms to increase employers’ vacancy views by 3–10× without degrading user relevance. Uses real-time, position-aware ranking adjustments instead of static boosts, and addresses issues like lack of a fixed position→click formula and the overheating/disappearance of boosted listings in high traffic environments.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:32Business ranking challenges
- 6:07A/B testing ranking
- 8:37Boost novelty and low views
- 9:39PID controller basics
- 11:06PID for ranking targets
- 13:52PID formula and score
- 14:36Ranking restrictions and elasticity
- 15:18Final ranking in practice
- 17:45Dynamic ranking behavior
- 19:10Balancing UX and business
- 20:34Architecture and data flow
- 22:16PID ranking advantages
- 23:22Coefficient optimization
- 25:49Key takeaways
- 27:40Q&A
related talks
11mTechTalk "Not by Time Spent Alone: How We Build a Ranking System"
39mRanking Listings with Machine Learning and ElasticSearch
23mML-Powered Search & Recommendation 101
38mLearning to Rank: where search meets machine learning #bbuzz
43mRecommendation Systems in Market Research
37mSearch and Recommendations: 3 Sides of the Same Coin #bbuzz
25mBetter search relevance using Learning to Rank at mobile.de
20mDataEngBytes 2023 - PER-T1-06 - Kamal Hossain