CISRI: A Crime Investigation System Using the Relative Importance of Information Spreaders in Networks Depicting Criminals Communications

In this paper, we propose a forensic analysis system called crime investigation system using the relative importance (CISRI) that helps forensic investigators determine the most influential members of a criminal group, who are related to known members of the group, for the purposes of investigation. In the CISRI framework, we describe the structural relationships between the members of a criminal group in terms of a graph. In such a graph, a node represents a member of a criminal group, an edge connecting two nodes represents the relationship between two members of the group, and the weight of an edge represents the degree of the relationship between those two members. Using this representation, we propose a method that determines the relative importance of nodes in a graph with respect to a given set of query nodes. Most current approaches that study relative importance determine the relative importance of a node under consideration by estimating the contribution of each query node individually to the importance of this node while overlooking the contribution of the query nodes collectively to the importance of the node under consideration. This may lead to results with low precision. CISRI overcomes this limitation by: 1) computing the contribution of the overall set of query nodes to the importance of a node under consideration and 2) adopting a tight constraint calculation that considers how much each query node contributes to the relative importance of a node under consideration. This leads to accurate identification of nodes in the graph that are important, in relation to the query nodes. In the framework of CISRI, a graph is constructed from mobile communication records (e.g., phone calls and messages), where a node represents a caller and the weight of an edge reflects the number of contacts between two callers. We evaluated the quality of CISRI by comparing it experimentally with three comparable methods. Our results showed marked improvement.