Make It Under 100 Milliseconds: Contextual Ads on Sphinx
summary
Describes Avito.ru’s contextual advertising system that must select a few relevant ads from millions of listings in real time. Explains using Sphinx as the search/index engine, applying millions of per-query weighting coefficients, and engineering optimizations and tricks to reduce end-to-end ad selection latency to under 100 milliseconds.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 1:06System requirements
- 3:57Click probability prediction
- 8:53Search engine choice
- 12:26Sphinx setup & ranking
- 20:38Optimizing for 100ms
- 25:15Performance results
- 27:01Q&A
related talks
47mWhere Your Ads Live
47mWhere Your Ads Live
44mAdvertising at the Speed of Light (DMP Platform)
01h 02mAchieving High Load in Advertising Technology
40mIn Search of... A better, fast
34mHow We Handle Over 1 Trillion HTTP Requests ...
46mDesign of the aviasales Metasearch Engine
41mThe evolution of search at Avito
