Shortlisting Influential Members of a Criminal Organization based on their Mobile Communication Data

Abstract

Hundreds of emails are sent every day around the world to trade and exchange data within every organization. Consequently, all of these interactions constitute a social network link. Social networks are a useful source of information about people's actions, and analyzing them might show groups of people that communicate in similar way. Emails and other records like mobile communication data are employed by digital forensic investigators to identify relationships between the members of criminal organization to construct criminal network and infer the most influential members of these communities. Those members play central role and usually targeted for removal or surveillance by the investigators. The key goal of this research is to see how centrality metrics features extracted from social network, when paired and enriched with content-based features, might reveal influential members within criminal organizations. As a method of addressing the research question, unsupervised clustering and topic modelling approach were employed. For clustering, K-means was evaluated and latent dirichlet allocation (LDA) for topic modelling task. Proposed approach allowed to compare betcween traditional methods of network analysis-based approach in comparison to network analysis-based approaches supported by machine learning performance. The results outperformed existing approaches.

Date of Award	Jun 2021
Original language	American English