Research on User Identity Authentication Based on Online Behavior Similarity

Abstract

AbstractNetwork security is not only related to social stability, but also an important guarantee for the digital intelligent society. However, in recent years, problems such as user account theft and information leakage have occurred frequently, which has greatly affected the security of users’ personal information and the public interest. Based on the massive user click behavior data and graph embedding technology, this paper proposes the Graph2Usersim (Gp2-US) to analyze the similarity between the user’s historical behavior characteristics and online behavior characteristics to accurately identify the user’s identity, and then distinguish the user’s abnormal online behavior. To be specific, firstly, based on the empirical click-stream data, the user’s historical behavior and online behavior are modeled as two attention flow networks respectively. Secondly, based on the Graph2vec method and drawing on the theory of molecular fingerprinting, the nodes of the attention flow network are characterized as atoms, and edges are characterized as chemical bonds, and the network is simplified using the structural features of compounds to generate feature vectors that can identify users’ historical and online behaviors. Finally, the behavior similarity algorithm Gp2-US proposed in this paper is used to accurately identify users. A large number of experiments show that the accuracy of the algorithm Gp2-US is much higher than that of the traditional algorithm. Based on 10 days of historical user behavior data, it can accurately identify its identity characteristics and accurately determine abnormal account behavior. The research conclusions of this paper have important theoretical value and practical significance in inferring abnormal user behaviors and monitoring public opinion.KeywordsAttention flow networkGraph embeddingSimilarity measurementFeature vectorIdentity authentication