A mathematical investigation on retrieval performance evaluation measures of information retrieval algorithm

Abstract

Retrieval performance evaluation measures of information retrieval algorithm are the foundation of evaluating and improving information retrieval algorithm. In this paper precision and recall are analyzed mathematically, and a theorem is proved: a good IR algorithm will make precision and recall to be maximal at the same time in an ideal vector space, then a new general evaluation measure based on this theorem is proposed to evaluate information retrieval algorithm.