Latent Inter-relation Augment between Crimes and Criminals using Factorization Machines

Abstract

In the process of criminal investigation, it is a very important to understand the relationship between the case and related persons. It is difficult and cumbersome to Figure out every time in the fragmented cases. To leverage this relational information, Graph-based machine learning-based approaches are being studied recently and natural in criminal. If you construct a crime network composed of criminal cases and related persons, a more systematic, in-depth, and broader investigation is possible by utilizing connection information. However, since previously known connection between cases and persons is within the range of a single case, it is very sparse, so it is hard to beyond a single case. In this study, we propose a method to infer the latent relationship using this rare case-person connection information and augment the existing network. Through this, it was confirmed that the augmented crime network can make maximum use of information beyond a single case, and the network expressive power is greatly improved for cases where single-case information was initially small. To verify the augmentation performance, it was verified by applying it to the actual crime data collected by the Korean National Police Agency.