Multi-Dimensional User Classification Modeling And Implement Based On Graph Neural Network Algorithm

The deep integration of social networks with daily life has made people's lives more and more convenient and interpersonal,and the inappropriate behavior of some spammers users can induce and mislead many Internet users who do not know the truth and lack the ability to discern.If they are not identified and controlled,they will pose a threat to the ecology of online public opinion and ideological security.With the widespread deployment of technologies related to social network analysis,such inappropriate behaviors have been curbed to a certain extent.However,the huge scale of users and complex user behavior in social networks make it difficult to distinguish undesirable users among them,and an effective and quick spammer detection is currently a major challenge for the industry.The following challenges exist in the identification of social network users.(1)Previous research usually divide user into simply two categories,spammers and normal users,with little deeper analysis of the concept of spammer itself,resulting in some users with suspected spammer behavior being in the blind spot of the research from the beginning.(2)The lack of available dataset make it difficult to meet the research needs.(3)The large scale of social networks and the fast dynamic changes require fast,efficient and flexible solution to handle the challenges.Sina Weibo,as one of the commonly used social network in China,can be used as a case study for user data analysis and can be effectively extended to other social network.Around the above challenges of social network user identification,the main research contents and main results of this paper are as follows.(1)To address the problem of insufficient awareness of the concept of spammer,this paper summarizes the typical features of several types of social network users through information mining of Sina Weibo.Then the paper proposes a three-dimensional model of anti-bot analysis-publicity-hashtag manipulation and constructs a feature set based on this model,which includes four new features,to better describe user's identity.According to the model,the paper defines social network users into four categories:harmless users,publicizing users,hashtag hijacking users,and malicious publicizing users.(2)To address the problem of lack of datasets,this paper creates two new Weibo user datasets through crawler collection and manual labeling,and reprocesses and labels an existing Weibo user dataset.And one more dataset is processed and created with the tag propagation algorithm.(3)To address the need for efficient algorithms,this paper constructs a dataset containing new features based on the proposed three-dimensional analysis model,and applies these new features to both traditional algorithms and graph neural network algorithms to conduct experiments.The experimental results show that the model and features proposed in this paper have good classification efficiency under various algorithms for the classification and bot behavior recognition tasks of Weibo users,and have better classification performance when run under the graph neural network algorithm GraphSAGE,which proves the usability of the proposed model.