Unverified50% confidenceFactExact time
The core metrics of approximate nearest neighbor (ANN) search are recall (recall@k), i.e., the proportion of the true Top-k hit by the approximate results, and the corresponding speedup (the multiple of QPS improvement relative to exact search).
1
Sources
50%
Confidence
Long-term
Relevance
11/5/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedIn engineering practice, recall (the proportion of the returned results that are true nearest neighbors) is usually used to measure the loss introduced by ANN approximate search.74% similarUnverifiedRecall is defined as the proportion of the returned Top-K results that truly belong to the set of true nearest neighbors.73% similarUnverifiedANN trades a small amount of recall for order-of-magnitude speed improvements, and its core metric is recall@K.73% similarUnverified在歧义点的四类行为中，'先搜再问'成功率最高，平均通过率达93.4%，远高于'直接猜'的56.5%，'搜多次再猜'为51.5%68% similarUnverified在从50%到92%的准确率提升中，REPL贡献最大（直接从50%跳至74%），随后领域知识注入、模糊搜索、公式追踪、系统prompt优化和bug修复层层累积66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1140482API
curl https://kongchang.com/api/v1/knowledge/claims/1140482MCP
get_claim(id=1140482)