Linking Multiple Online Identities in Criminal Investigations: A Spectral Co-Clustering Framework

Xiaohui Han, Lianhai Wang, Chaoran Cui, Jun Ma, Shuhui Zhang

Abstract: Online identities (OIDs) refer to accounts that Internet users create on Web platforms. In investigations of OID-involved criminal cases, investigators often encounter multiple OIDs used by the same criminal among a number of suspect OIDs. Recognizing such OIDs is referred to as OID linkage (OL) and is a critical task, because it helps investigators consolidate information from different sources, identify new investigation leads, perform further analysis by connecting seemingly unrelated cases, and finally ideate the real criminal. However, few existing OL techniques can achieve satisfactory performance in criminal investigation scenarios due to information asymmetry, information unreliability, and a lack of training data. We propose an unsupervised spectral co-clustering-based OL framework that takes OID access trajectories as linkage evidence. By leveraging a spectral co-clustering algorithm, we integrate access location consistency and access time consistency as heuristics to direct the OL process. Experiments performed using actual investigation data demonstrate the feasibility and promise of the proposed framework.