A new retrieval framework called WeSearch has been developed for WeChat Search, offering a unified, model-centric approach to handling large-scale, evolving content. Unlike previous methods that assign fixed Semantic IDs, WeSearch retrieves relevant clusters and then ranks individual documents, improving relevance modeling and enabling frequent model refreshes. This system has been fully deployed in WeChat Search, demonstrating significant improvements in click-through rate, per-user article clicks, and a substantial reduction in retrieval resource costs and latency. AI
IMPACT Enhances large-scale search relevance and efficiency, potentially influencing future search engine architectures.
RANK_REASON The item describes a research paper detailing a new retrieval framework and its deployment in a large-scale search engine.
Read on arXiv cs.IR (Information Retrieval) →
- alphaXiv
- CatalyzeX Code Finder for Papers
- Connected Papers
- CORE Recommender
- DagsHub
- Gotit.pub
- Hugging Face
- Influence Flower
- Litmaps
- ScienceCast
- scite Smart Citations
- Semantic IDs
- WeChat Search
- WeSearch
- click-through rate
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →