Research On Account Classification Based On Heterogeneous Network

As one of the most popular social systems today,Twitter has a huge number of users,and users frequently interact on Twitter every day.Therefore,the Twitter system contains a huge amount of information,and it has also received more and more attention and research from researchers.The purpose of this thesis is to classify accounts on Twitter.In the existing research,most of them only use the attribute characteristics of the account and ignore the network characteristics or only use the simple network characteristics.The network characteristics are very important for account classification.This thesis combines attribute features and complex network features to try to classify accounts.The main work of this thesis is summarized as follows:(1)Based on the Twitter social network,a Atributed Heterogeneous Information Net-work(AHIN)is established.The Twitter social network system contains a variety of dif-ferent types of nodes,such as accounts,tweets,events,etc.,which we call heterogeneous nodes.These different categories of nodes can be connected through different social re-lationships,such as users posting tweets,tweet-related events,and so on.In addition,the node itself also contains a large amount of characteristic attribute information,such as the account's homepage information.Based on these complex social relationships and node attributes,this page create a Atributed Heterogeneous Information Network.(2)Based on AHIN,this thesis proposes a network representation learning method called AHIN2 vec.Random jump is performed based on the meta-path and the similarity of attributes between nodes,and the jump track sequence text of the account is obtained.Then use the processing method in nlp to achieve vectorization of account nodes,and then use traditional machine learning methods to classify accounts.Experimental results prove that the method proposed in this thesis is better than other mainstream network representation learning methods at the current stage.(3)Based on the Graph Convolutional Network,an account classification method called multi-gcn is proposed.Because gcn is for homogeneous network graphs,various single-view graphs are extracted based on different single-view graphs from AHIN.Each single-view graph is processed with gcn,and then multiple networks are processed us-ing attention Combine to get the final classification result.Experiments show that the classification accuracy of this method is as high as 88 %.