hit rate
A retrieval evaluation metric that measures the proportion of queries for which at least one relevant document is retrieved. It is a binary judgment that ignores the ranking position of the hit, and in multi-document query scenarios it leans toward any-hit logic.
Timeline (last 90 days)
In RAG evaluation, the two rules 'any expected document in top-k' and 'all required evidence in top-k' may give completely opposite conclusions for multi-document queries, and the two are not interchangeable.
hit rate usually refers to 'the proportion of queries that hit at least one relevant document', is a binary judgment and does not consider the rank of the hit; recall@k measures the proportion of all relevant documents recalled in the Top-K, and is a continuous value.
In the single-document scenario, hit rate and recall@k trend consistently; under multi-document queries, hit rate is naturally biased toward any-hit logic, while recall@k calculated as 'number of relevant documents recalled / total number of relevant documents' is closer to the all-evidence perspective.
Many papers focus solely on the hit rate metric when comparing caching algorithms, while ignoring the CPU overhead, memory footprint, and implementation complexity of the algorithms themselves.
All Facts (4)
In RAG evaluation, the two rules 'any expected document in top-k' and 'all required evidence in top-k' may give completely opposite conclusions for multi-document queries, and the two are not interchangeable.
50%Unverifiedhit rate usually refers to 'the proportion of queries that hit at least one relevant document', is a binary judgment and does not consider the rank of the hit; recall@k measures the proportion of all relevant documents recalled in the Top-K, and is a continuous value.
50%UnverifiedIn the single-document scenario, hit rate and recall@k trend consistently; under multi-document queries, hit rate is naturally biased toward any-hit logic, while recall@k calculated as 'number of relevant documents recalled / total number of relevant documents' is closer to the all-evidence perspective.
50%UnverifiedMany papers focus solely on the hit rate metric when comparing caching algorithms, while ignoring the CPU overhead, memory footprint, and implementation complexity of the algorithms themselves.
50%