Reproduce “Personalized Query Auto-Completion for Long and Short-Term Interests with Adaptive Detoxification Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhibo Wang, Xiaoze Jiang, Zhiheng Qin, Enyun Yu, Han Li
The LaD model enhances query auto-completion by integrating hierarchical modeling of users' long-term and short-term interests with an adaptive detoxification mechanism. Deployed on Kuaishou search, the system achieved a +4.08% increase in Click-through Rate and a 0.81% reduction in problematic completions through human evaluation.
Key insights
- Hierarchical modeling of user interests, distinguishing between coarse-grained long-term and fine-grained short-term preferences, significantly enhances the personalization quality of QAC systems.
- The novel Reject Preference Optimization (RPO) with online `[Reject]` token injection effectively enables adaptive detoxification, allowing the model to selectively suppress highly sensitive suggestions without sacrificing relevance for benign queries.
- Joint training of personalization and adaptive detoxification within a single generative framework achieves a superior balance between query relevance, generation quality, and content safety in real-time industrial settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.