An Enhanced Group Abnormity Detection Model in Social Networks Through Multi-Scale Knowledge Graph-Based Deep Learning

Abstract

In area of group abnormity detection, deep learning-based methods have received more attention in recent years. But existing research works mostly failed to well-integrate complex graph-level characteristics of social network context into feature representation. To deal with this issue, this paper proposes an enhanced group abnormity detection model in social networks through multi-scale knowledge graph-based deep learning. First, multi-scale knowledge graph is formulated to extract and field knowledge as the fundamental support. Then, a recurrent neural network structure is utilized to express graph-level features, so as to aggregate the information of the surrounding nodes. For output of group abnormity detection results, the interaction information among users is further considered, and the correlation analysis of abnormal behaviors is conducted based on the probability expression. On this basis, probabilistic soft logic is used to express the relationship between abnormal association users and their behavior occurrence. Finally, a set of identification rules are developed to discover abnormal association users, and those who have the same association probability value are identified as abnormal user groups. Experiments are conducted on social network dataset to evaluate performance of the proposed model. The results reveal that it can make group abnormity identification with higher accuracy compared with baseline methods.