Mining The Association Of Multiple Virtual Identities Based On Multi-Agent Interaction

Abuses of online anonymity make identity tracing a critical problem in cybercrime investigation. To solve this problem, this paper focuses on the feature of authors' behavior in time slices and tries to mine the association of multiple virtual identities based on multi-agent interaction. We propose the recognition model MVIA-K based on knowledge management. In MVIA-K, agents perform distributed mining to get candidate author groups as local knowledge in each time slice. Then high-quality knowledge is extracted from the local knowledge and used as priori knowledge to guide other agents' mining process. Finally distributed knowledge is integrated on the basis of knowledge scale. Experiment with real-world dataset shows that MVIA-K has a very promising performance, which can filter the noise data effectively and outperform Author Topic model.