• Computer Science > Information Retrieval [Submitted on 17 Dec 2025] Title:When & How to Write for Personalized Demand-aware Query Rewriting in Video Search View PDF HTML (experimental)Abstract:In video search systems, user historical behaviors provide rich context for identifying search intent and resolving ambiguity. • However, traditional methods utilizing implicit history features often suffer from signal dilution and delayed feedback. • To address these challenges, we propose WeWrite, a novel Personalized Demand-aware Query Rewriting framework. • Specifically, WeWrite tackles three key challenges: (1) When to Write: An automated posterior-based mining strategy extracts high-quality samples from user logs, identifying scenarios where personalization is strictly necessary; (2) How to Write: A hybrid training paradigm combines Supervised Fine-Tuning (SFT) with Group Relative Policy Optimization (GRPO) to align the LLM’s output style with the retrieval system; (3) Deployment: A parallel “Fake Recall” architecture ensures low latency. • Online A/B testing on a large-scale video platform demonstrates that WeWrite improves the Click-Through Video Volume (VV$>$10s) by 1.07% and reduces the Query Reformulation Rate by 2.97%. • Current browse context: cs.IR References & Citations export BibTeX citation Loading…

Article Summaries:

  • A new framework, WeWrite, tackles personalization in video search by leveraging users’ historical behavior to better infer intent and reduce ambiguity. The system first determines when personalization is needed through a posterior‑based mining strategy that selects high‑quality log samples. It then trains a large language model using a hybrid supervised fine‑tuning and Group Relative Policy Optimization to align query rewrites with the retrieval engine. For deployment, a parallel “Fake Recall” architecture keeps latency low. In live A/B tests on a major video platform, WeWrite raised click‑through volume for videos longer than 10 seconds by 1.07 % and cut query reformulation rates by 2.97 %.

Sources: