Identifying Criminals’ Interactive Behavior and Social Relations Through Data Mining on Call Detail Records

Abstract

A new method of identifying criminals’ interactive behavior and social relations based on graph computation and clustering model is proposed in this paper. Firstly, a graph database is created in Neo4j to generate a knowledge graph on the basis of calling and crime datasets, and graph traversal and algorithms are implemented to identify criminals’ interactive patterns and practical clues for detecting suspects. Secondly, nine typical features are extracted from calling records. Thirdly, a Gaussian mixture model based on the features is built using Python, and the distributed computing of the model is achieved in Spark. The experiments prove the graph traversal and algorithms can reveal how criminals contact with each other, and the Gaussian mixture model can identify 5 kinds of calling patterns. Furthermore, this study finds that two criminals usually keep a special interactive pattern, in which duration of keeping in contact and holding time are long, the amount of calls is large and the calls happen in the wee hours mostly.