Research On User Identity Matching Algorithms Across Social Networks

The purpose of cross social network user identity matching is to detect whether users from different social networks belong to the same person.Information in these social networking sites is often of great importance in other fields,such as information diffusion and link prediction,cross domain recommendation,authentication and privacy protection.At present,cross-social network user identity matching algorithms mainly use personal attribute characteristics,network structure characteristics and behavior characteristics for research.Although the existing research has made some progress,there are still some shortcomings.For example,the existing research fails to fully explore the closeness between users.However,it lacks a large number of accurate and reliable label users,and fails to effectively use the global network structure characteristics and local network structure characteristics.Therefore,in response to the above mentioned shortcomings,this paper has done a further research and proposed two user identity matching algorithms.The specific research contents are as follows:(1)Aiming at the problem of ignoring the closeness between friends in existing methods,a friend closeness based user identity matching algorithm(FCUM)is proposed.This algorithm is a semi-supervised cross social network user identity matching algorithm.Firstly,the structure of user friends obtained from real social network is extended,and then the user vector space is obtained by using the derived Skip-gram model.Then attention mechanism is used to quantify the closeness of friends,and then generate the vector space of close friends.Through the joint optimization of user individual similarity and close friend similarity in a single objective function,the user individual similarity and close friend similarity are fully utilized.In addition,a bi-directional matching strategy is designed to solve the problem of high cost of manual label users.The experiment on public data sets shows that the proposed algorithm is better than other methods which only consider the similarity of users.(2)Aiming at the problem of insufficient utilization of local network structural features by existing methods,a multi-level friend structure based user identity matching algorithm(MLUM)is proposed.Firstly,the algorithm mines the multi-level friend structure of users,and generates low-dimensional vectors that can represent users in social networks.The siamese network based on autoencoder is designed to obtain more efficient user vectors.Then the algorithm constructs a dynamic strategy selection mechanism to enhance the overall prediction performance of the model.Finally,the algorithm uses the global and local network structure features through the co-training framework to enhance the credibility of new label users.Experiments on public data sets show that the proposed algorithm improves the matching accuracy compared with several existing methods based on user network structure.