Tencent's WeSearch fully deployed in WeChat Search: -41% retrieval cost, +0.68% CTR
_reachsumit · x · 2026-11-05
Tencent researchers present WeSearch, a unified model-centric retrieval framework now fully deployed as the primary semantic retrieval path in WeChat Search:
- Prevailing methods assign fixed hierarchical Semantic IDs, assuming each doc follows one semantic path — which breaks on multi-topic content. WeSearch instead retrieves query-relevant clusters, then ranks atomic documents within that candidate space, decoupling pruning from fine-grained relevance modeling.
- With relevance-aware training, compact doc representations, and incremental index updates, it supports weekly (or daily) model refreshes.
- A one-month full-traffic A/B test shows +0.68% CTR, +2.27% per-user article clicks, 41% lower retrieval resource cost, and 19ms faster graphic-text retrieval over corpora of hundreds of millions to billions of docs.
More from Research
- Claude drifts from 🌀 to 💖: refined AI symbols become less legible to humans (67%→44%) — mimi10v3 · 2026-11-05
- Engram: brief training fixes long-run models' increasingly inscrutable notes — scott_linderman · 2026-11-05
- JITEngine synthesizes custom LLM inference engines, 2.7x faster on Apple Silicon — CShorten30 · 2026-11-05
- NMSLIB author publicly slams new vector search preprint for downplaying prior work — srchvrs · 2026-11-05
- NMSLIB author: new preprint cites us to back an unsubstantiated claim — srchvrs · 2026-11-05
- Foresight-SWE benchmark: coding agents solve 60.7% of issues but only 14.8% without introducing bugs — OfirPress · 2026-11-05