Ranking chat logs for crime investigation

Abstract

Aiming at the needs of criminal investigators to quickly find chat logs related to crime case, a ranking method for chat texts has been proposed. First, query keywords were used to search chat texts, and the feedback texts were divided into relevant texts and irrelevant texts with manual annotation, then non-negative matrix factorization (NMF) was utilized to obtain the implicit semantic relationship of chat texts, finally chat logs could be ranked based on the scores calculated by similarity. Experiments show that the method proposed in this paper can quickly retrieve and get the messages of interest in a lot of chat logs, which can facilitate crime investigation.