Research On User Alignment Technology Based On Joint Embedding Of Heterogeneous Networks

With the rapid development of the Internet,social network has become more and more important in People's Daily life.People usually register multiple different social network accounts to enjoy the functions of different social networks.User alignment refers to aligning different accounts belonging to the same natural person in the real world in multiple social networks,so as to integrate the fragmented information of users scattered in multiple networks,which is of great significance to network security,data mining of social networks,product recommendation and other fields.At present,there are still defects in user alignment.First,many user alignment studies embed social networks into vector space alone,which results in the loss of network information and the inability to effectively dig hidden information between networks.Second,in the use of user attributes,most user alignment tasks use vector splicing,which cannot effectively combine network structure with attributes.Aiming at the above problems,this thesis adopts the method of heterogeneous network joint embedding to carry out the research on user alignment.The main research is as follows:1.A user alignment algorithm based on knowledge graph joint embedding is proposed.First,the seed anchor user pair set is used to increase the number of positive examples and make the seed anchor user pair closer in the embedded space;Secondly,the Near?K negative sampling method is proposed to ensure the negative sample quality and improve the embedded space quality;Then,the structure similarity is proposed,which fully considers the similarity of users on the network structure,and measures the user similarity together with the cosine similarity;Finally,the greedy matching method based on adaptive threshold is used to find the potential anchor user pairs,and the newly found anchor user pairs in each iteration are added to the next round of training,so as to optimize the embedding space and avoid error passing.Experimental results show that the performance of the proposed algorithm on real social network datasets is greatly improved over the benchmark algorithm,and it can effectively solve the problem of cross network user alignment.2.Applying graph convolutional neural network to user alignment task,a user alignment algorithm based on graph convolution neural network is proposed.Firstly,a user intimacy measurement method is proposed based on the jackard coefficient,so that the intimacy between users in the social network can be reflected in the adjacency matrix,which can improve the alignment performance.Secondly,the algorithm proposes the adjacency matrix construction method,so that the graph convolutional neural network can be applied to the cross network alignment task.Finally,the algorithm combines the user name attributes with high authenticity and easy access with the network structure,and acts on the user alignment task together to further improve the alignment performance.The experimental results show that the algorithm effectively integrates the social network structure and user attributes,and improves the alignment performance of cross-network users.