Recommendation Systems in Market Research
summary
Platform-level recommender replaces brittle rules with a production factorization approach: factorization machines via xlearn, trained on user survey attempts and combined with a multi-criteria optimization that weights payout duration and conversion. Model is retrained every two minutes to keep new users and surveys current, serves predictions in ~70 ms prediction latency, and handled 140 million requests last week. Results show conversion uplift (roughly 20–40% in impacted cohorts) and higher earnings-per-click versus competitors like Lucid.
AI-written summary · about
chapters (18)
- 0:00Recommender systems intro
- 1:15Matching users with surveys
- 5:09Personalized survey matching
- 6:34Recommender system types
- 10:07Collaborative filtering challenges
- 13:39Factorization machines
- 16:12Model training & prediction
- 18:20Real-time recommendation infra
- 22:55Impact on satisfaction & revenue
- 24:03Q&A on model testing
- 28:43Cold start problems
- 31:12Revenue optimization
- 32:59Content-based systems
- 34:05Initial recommendations
- 35:50Model performance trends
- 38:01Sampling trade-offs
- 40:31Behavioral data & ads
- 42:20Conclusion
related talks
41mRecommender Systems: Crafting Effective Personalised Experiences
58mRecommender Systems – A Walkthrough
21mRecsys Keynote: Improving Recommendation Systems & Search in the Age of LLMs
34mRecommender Systems
32mLeveraging recommender systems to personalise search results
32mLet's talk about Recommendation Systems!
23mML-Powered Search & Recommendation 101
52mReinforcement Learning in Recommender Systems: Some Challenges