WeFair: Fidelity-Aware Alignment from Human Feedback for Personalized Query Rewriting in Short-Video Search

Key points

• Short-video search requires compact query rewriters that infer personalized intent from brief, ambiguous queries under strict latency constraints.1
• We first mine actual human reformulations from search sessions, then apply multidimensional LLM screening and contextual verification to filter unsupported transitions without extensive manual annotation.1
• In the final on-policy optimization stage, we construct a strict verification reward based on actual human reformulations to further boost performance with group relative policy optimization.1
• Online A/B testing in WeChat short-video search shows a 1.576% relative increase in effective clicked-query rate (VV>10s) and a 1.731% relative reduction in query reformulation rate.1

Sources (1)

• [1]WeFair: Fidelity-Aware Alignment from Human Feedback for Personalized Query Rewriting in Short-Video Search
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 12, 03:10 PM

  To build an effective rewriter that remains faithful to user intent, we propose WeFair, Fidelity-Aware Alignment from Human Feedback for Personalized Query Rewriting in WeChat short-video search.

  Short-video search requires compact query rewriters that infer personalized intent from brief, ambiguous queries under strict latency constraints.

Extractive summary: sentences quoted from the sources.