Research On Classification Method Of Social Media Accounts Based On Heterogeneous Information Fusion

With the rapid development of network technology,information can spread rapidly around the world,and social media has gradually become an important way for the public to obtain information,communicate and interact.Due to the large number of social media users,a large amount of data is generated every day.However,most of the information in these data is of no value,making it difficult for users to obtain useful information from it,and reducing the efficiency of information use.Social media user account classification is the process of identifying accounts with common characteristics from massive amounts of information.By categorizing account numbers,not only can the efficiency of information acquisition be effectively improved,but the acquired information can also be applied to practical applications such as recommendation systems and question and answer systems.Existing social media account classification methods generally use attribute information or text information of accounts to construct features,and use supervised learning methods to classify accounts.However,in practical applications,the existing account classification methods still have the following problems: Accounts have a variety of information,but because actual social media has the characteristics of high noise and sparseness,a single information cannot be fully described.Existing methods can only choose one or a few types of information due to problems such as computational complexity.They neither give more thought to the fusion process of multiple types of information nor consider the implicit relationship between multiple types of information,resulting in The loss of information leads to a mediocre effect of final account classification.In response to the above problems,this article conducts research on the classification of social media accounts.The main contributions are summarized in the following two aspects:(1)Aiming at the problem that the existing methods do not make full use of account information,this thesis proposes an account classification method based on multi-modal fusion.This method extracts multi-modal features on the basis of comprehensive consideration of account attributes,texts,and social relationships between accounts,and uses tensors to fuse these features.Compared with the traditional method of only using partial account information for account classification,the multi-modal feature fusion method proposed in this thesis can comprehensively consider all the information of the account and obtain a better classification effect.Through experiments,the accuracy of the method in this thesis has reached 93.74%.(2)In order to further explore the connection between account information and improve the effect of account classification,this thesis proposes a method of constructing a heterogeneous graph convolutional attention network to classify accounts.Based on the convolutional neural network GCN,this method introduces a heterogeneous information network model and adds an attention mechanism to enable it to be applied to heterogeneous networks and assign weights according to node types.Using this method can fully mine the account information itself and its implicit relationship.Through experiments,the accuracy of the method in this thesis has reached 96.6%,which is higher than the previous traditional account classification methods,which proves the effectiveness of the method in this thesis.