Fighting criminals: Adaptive inferring and choosing the next investigative objects in the criminal network

Abstract

When a criminal probabilistic network has been constructed, criminal investigators can access verified information about some nodes of network after investigations. Effective and efficient techniques are needed to help law enforcement and intelligence agencies to infer the state of other nodes and choose the next new investigative objects in the criminal network. In this paper, we propose a technique that employs a belief propagation algorithm to help criminal investigators to infer the criminal probability of other members by using verified partial information. In an updated criminal probabilistic network, this paper also presents a technique called EMPFS which extends the modified PFS algorithm. EMPFS algorithm is used to solve choice of next key investigative objects from criminal network. Experimental results show that the precision and efficiency of two techniques might be improved by exact constructing of the crime probabilistic network.

Introduction

Internet newsgroups, bbs, and chartrooms are an appealing way for members of such groups to communicate because they are easily accessed from almost anywhere in the world and offer the potential for camouflaging organized crime among the numerous background communications. Members of organized crime networks are often camouflaged as legal Internet users. These users in the criminal networks (people, IP-address, etc.) are usually modeled by nodes of graphs. The connection relations (trust or dependent relation) between members are represented by the graph edges. Generally, a criminal network is composed of criminals and legal people. When probability is used to describe the criminal probability of each member and dependent or trust relation between two members in the criminal network, the criminal network becomes the criminal probabilistic network.

Since members in the criminal network may be criminals or legal people, crime investigators often use probability to evaluate whether the member is a criminal or a legal person. The criminal probability of each member is incarnated by the “belief” of the criminal investigator about whether the member is a criminal. When several members in the criminal network are verified to have committed criminal activities, the crime investigator can get partial information about the criminal network. Crime investigators hope to use partial information to do the following tasks: (1) to re-evaluate the criminal probability of other members in the criminal network by using partial information and (2) to choose the next investigative objects for crime investigators in the newly evaluated criminal network. Both tasks are very time-consuming and labor-intensive. To finish the above two tasks, investigators may spend a large amount of time performing extensive database searches, reading crime reports, looking for clues of crime associations, and individually inferring the crime probability of other members of the criminal network.

Some technologies [1], [2], [3], [10], [11], [12], [13] have achieved in analyzing criminal networks; for example COPLINK techniques [1], [2], [3], [16], [17], [18], [19] have been applied to identify associations in criminal networks. Also, some discrete or continuous Markov models are used in statistical analysis to predict future crime events or processes and to identify hidden groups in a criminal network [4], [5]. SNA [6], [7] is used to measure individual centrality in the network and clustering, centrality measures, block modeling, and multidimensional scaling (MDS) approaches from SNA is employed by Xu and Chen [8], [9] to study criminal networks based on crime incident data. However, these techniques cannot provide adequate solutions for the above two tasks.

To solve the above two tasks, we propose using belief propagation to update the beliefs of crime investigators about the criminal probability of other members in the criminal network using partial information. Belief propagation uses Bayesian rules and maintains an association probability distribution over each member (i.e., each node in the criminal network graph). In the newly updated criminal network, we propose using an extended MPFS (modified PFS) algorithm to choose some member as the next investigative object of crime investigators by using partial prior information. These investigative objects can help crime investigators to analyze criminal network at the lowest cost.

The rest of the paper is organized as follows. In Section 2, we present adaptive inferring and propose the ‘choosing next objects’ model. Section 3 presents the concept of updating criminal networks based on belief propagation. Modified BFS is introduced and Extended MPFS is presented in Section 4. Case study and experiments are presented in Section 5, and conclusions are included in final section.