CAKGC: A Clustering Method of Cybercrime Assets Knowledge Graph Based on Feature Fusion

Illegal gambling and pornography are becoming increasingly rampant in cyberspace over the last ten years. Existing research of cybercrime governance mainly relies on detecting illegal websites and using simple rules to determine organizational affiliation. These methods are not only inaccurate enough, but also ignore other cybercrime assets and their complex relationships. To address these problems, we propose a novel cybercrime assets knowledge graph clustering method (CAKGC) based on feature fusion which combines heterogeneous node attributes with graph structure features of knowledge graph. We carefully analyzed different kinds of multi-source heterogeneous cybercrime assets exposed on the Internet and their intricate relationships, providing preparation for designing ontology and constructing a comprehensive cybercrime asset knowledge graph. Moreover, two features extraction strategies are adopted to learn heterogeneous node attributes and graph structure features automatically. Finally, we fuse two-level features by dimensionality reduction and apply clustering algorithms to discover highly dense cybercrime assets of groups. Experimental results on real-world cybercrime datasets demonstrate the superiority of CAKGC in terms of clustering accuracy (ACC) and normalized mutual information (NMI) and purity, outperforming advanced baseline methods.