Search, ads, and recommendation¶
简体中文 · English
Follow the path from finding candidates to ranking and serving: semantic representations, user knowledge, long histories, generative recommendation, feedback, and deployment.
Choose a subtopic and start with its paper questions.
Candidate generation and search matching
Can a stronger fine ranker recover relevant content that retrieval missed?
Semantic representations, user knowledge, and cold start
How does a new item without clicks enter a behavior-based recommender?
User histories and ranking backbones
Should compute go to history–history, candidate–history, or heterogeneous field interactions?
Generative retrieval and reasoning-based recommendation
What changes when items are represented as tokens?
Learning signals, multitask learning, and feedback bias
Which probability or value can observed clicks and conversions supervise?
Slates and business control: reranking, bidding, and long-term value
Why is sorting items by individual score not necessarily a good slate?
Automated research and efficient serving
When is a better offline method usable under real requests and continued iteration?