Multi-Armed Bandits in Recommendations
summary
Description of using multi-armed bandit algorithms in Avito’s Recommendations team as an alternative to A/B testing: technologies include Data Science and microservices; the problem is running many concurrent hypothesis tests and automatically discarding poor variants; the approach is to deploy multi-armed bandit methods to optimize reward from the start of an experiment and perform automatic selection.
AI-written summary · about
chapters (9)
- 0:00Introduction to recommendations
- 3:38Similar ads problem
- 11:08Testing config problems
- 13:38What are bandits
- 18:17Bandit strategies compared
- 22:14Target actions and seasonality
- 24:00Bandit results
- 24:42System architecture
- 26:29Q&A
related talks
21mLeveraging Multi-Armed Bandit Algorithms for Dynamic Decision Making
27mEnhancing Personalization Using Recommendation Systems & Multi-Armed Bandits
17mByron Galbraith, Chief Data Scientist, Talla, NYC 2017
22mPersonalizing Explainable Recommendations with Multi-objective Contextual Bandits
57mSearch Relevance: A/B testing to Reinforcement Learning
30mPractical Multi Armed Bandits
45mContextual Multi-Arm Bandit and its applications to digital experiments
15mAI-Driven Application Optimization