Research On Sockpuppet Recognition Method And Application Based On Heterogeneous Multi-source Features

In recent years,social networks have developed rapidly and are closely connected with people's lives.People share their daily lives with others through social networks,but there are also some negative effects,such as creating malicious accounts to pretend to be others,manipulating public opinion,disrupting the normal operation of social platforms,etc.Sockpuppet detection refers to the identification of different accounts created by the same user in the same social network,thereby banning the created malicious accounts,which is of great significance to the fields of social platforms and network security.Nowadays,there are still some shortcomings in the recognition of social network sockpuppet:(1)The problem of feature selection.At this stage,many sockpuppet detection methods manually select features and select the best combination of features.(2)The combination of network structure is not used in the research of sockpuppet detection.In response to the above problems,this thesis adopts the method of multi-source feature fusion to study the sockpuppet account recognition.The main research work of this thesis is as follows:1.Propose a sockpuppet detection algorithm based on adaptive multi-source feature fusion.First,extract the user's verbal features and non-verbal features,where the word embedding model is used for the content of the text,and the text is converted into word vectors,which are then spliced into sentence vectors;second,the two features are jointly embedded into a unified low-dimensional Fusion in the space to obtain a more comprehensive analysis of the text language characteristics and behavior characteristics of the sockpuppet account;then use the feature selection technology to adaptively select the optimal feature combination,compared to the greedy matching selection algorithm,which reduces the algorithm time Complexity;Finally,this article converts the sockpuppet problem into a classification problem,which can be classified using a variety of machine learning algorithms.The experimental results show that the method proposed in this thesis has a greater performance improvement than the benchmark algorithm,and can effectively identify social network sockpuppet accounts.2.The graph attention network is applied to the sockpuppet detection task,and a sockpuppet detection algorithm based on the graph attention network is proposed.First,use the Tanimoto coefficient to calculate the adjacency matrix between social network users,so that the graph attention network can be applied to the social network sockpuppet recognition task;secondly,the user attribute matrix is constructed to improve the accuracy of recognition;finally,the social network graph structure Combine with user attributes to further improve recognition performance.The experimental results show that the algorithm can effectively identify sockpuppet accounts in social networks.