Unverified50% confidenceTradeoffExact time
BM25 is a probabilistic improvement over TF-IDF that comprehensively considers term frequency, inverse document frequency, and document length normalization, but is unable to capture semantic similarity.
1
Sources
50%
Confidence
Long-term
Relevance
9/8/2026
First Seen
Sources
Related Entities
Related Claims
Unverified在法律和金融领域BM25依然强悍是因为这些领域高度依赖精确的专业术语匹配，向量嵌入模型无法精确区分违约与不构成违约这类语义相近却截然相反的表述68% similarUnverifiedBM25 solves TF-IDF's problem of inflated term frequency in long documents by compressing a single term's weight in longer documents67% similarVerified混合精度训练将权重保存为FP32，而前向和反向传播计算使用FP16或BF16，可将显存占用减少近一半61% similarVerifiedHybrid retrieval (combining dense vector retrieval with BM25 sparse retrieval) performs better than either method alone on most benchmarks.61% similarUnverifiedLLM 在润色时倾向于把简洁精准的技术表述改写得更华丽，反而引入模糊性，可能将有特定含义的技术术语优化成同义词导致精确性丢失59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/877692API
curl https://kongchang.com/api/v1/knowledge/claims/877692MCP
get_claim(id=877692)