Research And Implementation Of User Identity Technology Based On Multidimensional Embedding

Online social networks provide people with media platforms for communication and information sharing.Since different online social networks provide different services for people,in order to use different social media platforms,users generally have multiple online social media accounts.At this time,the data in online social network is used to identify the user's identity from multiple dimensions,and then the cross social media user alignment operation is carried out to realize the identification of social network user's identity in the network knowledge map,which is of great significance of the development of network security,recommendation system and other fields.Existing user identification methods do not make full use of user characteristic data,and the identification effect of users is difficult to reflect.In view of the above problems,this thesis focuses on how to integrate social network multi-source user information to achieve representation of users,and at the same time,align users across social networks based on user identification,iterate the representation of user identity,and user representation evaluation indicators.The main research work of the paper is as follows:(1).Based on the characteristics of social network user data,this thesis studies a user identification method NAT based on three dimensions of user network topology information,user attribute information and user text information.The information at three latitudes is represented by three independent vectors,including net embedding generated by graph embedding,attribute embedding generated by feature extraction and coding of user attribute information,and text embedding generated by Doc2 vec.Using link prediction as the evaluation indicator of the identification of user identity,based on the user vector representation obtained by the NAT method,comparing dozens of link prediction algorithms according to the cosine similarity of user vector,can achieve better AUC performance,which proves the effectiveness of user identification method NAT.(2).With the help of pix2 pix model in the field of image translation,we can train pix2 pix alignment model across social networks,use BiLSTM model with attention mechanism as the generator of user representation translation between social networks,and use multilayer BiLSTM model with attention mechanism as the discriminator,it trains the two-way translation model of user representation between social networks.In the experimental stage,through the comparison of various algorithms,the effectiveness of the anti user translation model is verified,which can effectively translate the social network user representation to another social network,identify and align the same users across the social networks.The aligned network knowledge map,which contains more user information,can further effectively iterate and improve the effect of identification of user identity.The experimental results show that the representation method based on multidimensional user information in this paper is effective,which can identify the user identity of social network.At the same time,it implements the translation alignment model across social network users,and provides a solution for the iterative representation of network knowledge graph users.