A topic network model to detect criminals without prior information

Crime analysis has been widely studied, but problem of identifying conspirators through topic network is still not well resolved. In this paper, we proposed a fuzzy clustering algorithm and a hard clustering method to detect criminals hidden in their talks, and both two algorithms took no use of individuals' prior identity information. Firstly, we built up an iterative formula to compute each node's local suspicion, and then we employed the fuzzy c-means (FCM) clustering algorithm and a new hard clustering method with noise (HCN) recently published in Science to get nodes' global suspicions, of which the values are used for conspirator identification. Experiments on a company's email dataset showed both algorithms obtained good identification results: known suspects gained relative higher values in FCM and were all assigned to the criminal cluster in HCN, known innocents got relative lower values in FCM and were all assigned to the innocent group in HCN. We also found these two algorithms were in good agreements, specifically, the HCN results helped determine the discrimination line of the FCM results and the FCM results could quantify the suspicion of each individual assigned in the criminal group of HCN. Copyright © 2015 American Scientific Publishers.