zilliztech / VectorDBBench

A Benchmark Tool for VectorDB
MIT License
455 stars 109 forks source link

Recall or Precision #305

Closed adbmal closed 2 months ago

adbmal commented 2 months ago

whether Vectordb has precision metrics? I've noticed that everyone is discussing recall, and in VectorDBBench, recall is used as a key metric. After looking into the calculation method of recall in VectorDBBench, I feel it should be referred to as precision. https://github.com/zilliztech/VectorDBBench/blob/296683255af8c0ba0b82bf60e96428cfbb58df2f/vectordb_bench/metric.py#L56-L62

https://en.wikipedia.org/wiki/Precision_and_recall

image

This issue has been bothering me for quite some time.

alwayslove2013 commented 2 months ago

@adbmal In fact, for top-k search results, our ground truth also consists of only k items. In this scenario, recall and precision are equal.