How to predict crime — informatics-inspired approach from link prediction

Abstract

Many social complex networks are best modeled as a bipartite graph and they evolve during time, thus, predicting links that will appear in them have become highly relevant and critical. Link Prediction is a key direction in social complex network research refers to estimating the possibility of the existence of non-existent links between node pairs. In criminal networks, link prediction can provide a very efficient way in the discovery of missing or hidden links and the detection of the underground groups of criminals. Only few works address the bipartite case, though, despite its high practical interest and the specific challenges it raises. Likewise, most of prior algorithms of link prediction consider a threshold. However, it is difficult to set such a proper threshold in advance for a given dataset. Hence, in this paper, we propose a new method called Latent Link Prediction based on Internal and Local Links (LLPIL) for bipartite networks. LLPIL is based on new proposed topological metric named reliability that can reflect the likelihood of two nodes to be connected. We exploit the proposed model to identifying and preventing future criminal activities. Extensive simulations show that our proposed algorithm has high prediction accuracy and low time complexity.