Verified75% confidenceBenchmarkExact time
Hybrid retrieval (combining dense vector retrieval with BM25 sparse retrieval) performs better than either method alone on most benchmarks.
3
Sources
75%
Confidence
Long-term
Relevance
7/16/2026
First Seen
Sources
Related Claims
UnverifiedHybrid retrieval, by fusing the results of dense vector retrieval (Embedding) and sparse retrieval (BM25/TF-IDF), can significantly improve retrieval recall.77% similarUnverified优质Prompt在大多数场景下比微调更高效，在数据量不足或需求模糊的阶段贸然做微调往往事倍功半67% similarUnverifiedIn scenarios with small context sizes, BM25 term-frequency retrieval outperforms semantic embedding solutions in terms of latency and cost66% similarVerifiedANN algorithms (such as HNSW and IVF-PQ) trade a very small amount of precision for orders-of-magnitude improvements in retrieval speed, and are the standard choice for production-grade vector retrieval65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/527110API
curl https://kongchang.com/api/v1/knowledge/claims/527110MCP
get_claim(id=527110)