Ranking factors architecture in Ozon search runtime
summary
Describes Ozon’s runtime ranking architecture that uses models and feature factors for search, including how ranking models are served (inference) in the search pipeline and how new models are promoted to production. Explains computing the dependency graph between models and their input features and the automated performance and quality checks run before deployment.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:12Ozon search ranking factors
- 8:00Ranking challenges
- 10:25Faster architectures
- 17:51Graph-based ranking
- 22:32Feature store role
- 26:05Formula generator pipeline
- 31:22Results and TTM reduction
- 32:26Key takeaways
- 35:16Q&A
related talks
11mTechTalk "Not by Time Spent Alone: How We Build a Ranking System"
37mSearch and Recommendations: 3 Sides of the Same Coin #bbuzz
47mHow We Increased Image Search Throughput
20mArchitecture of relevancy search at mobile.de
39mRanking Listings with Machine Learning and ElasticSearch
47mLearning to rank search results
38mLearning to Rank: where search meets machine learning #bbuzz
47mWhy, What, When and How of Running Diff Types of Sear...