Research On User Identification Algorithms Across Multiple Online Social Networks

As an important part of online social networks research,user identification across social networks has been playing an important role in cyberspace security management,personal service recommendation,social network analysis and etc.The study of user identification across social networks has made great process and many algorithms have been proposed,however,there are still some problems unsolved:(1)The user identification algorithms based on the topological information only focus on ego-network topological environment,and has ignored hidden labeled nodes,which are important for improving user node identification degree.(2)Present user identification algorithms based on the public profile attribute information adopt subjective weight-directed objective weighting method,which ignores each attribute's special meaning and function.(3)The user identification algorithms based on activity information analyze users' unique writing and reading styles,model user's behavior patterns,but ignore the dynamic property of activity information's evolution in the social networks.To solve the problems above,three kinds of user identification algorithms were proposed to further improve the user identification algorithm's accuracy and versatility.The main research contents are listed as follows:1.A hidden label nodes mining based user identification across multiple online social networks algorithm was proposed.By adding unmatched nodes with community clustering information,the proposed algorithm firstly mined hidden label nodes and put them into unmatched nodes' ego networks.Then it took advantage of the potential relationship information to improve the identification degree of the nodes to be matched,and chose perfect matching pair with label nodes.Finally,it conducted iterative operation to identify all nodes in the whole network.Experimental results on both synthetic random networks and online social networks show that,when compared with Ego-UI algorithms,the proposed algorithm can increase the recall and the F-1 measurement under the premise of ensuring the accuracy of user identification,and then can identify more accounts.2.A profile attributes information entropy weights determination based user identification across multiple online social networks algorithm was proposed.Firstly,the proposed algorithm specifically analyzed the data types and physical meanings of different attributes,and adopted different similarity calculation methods correspondingly;Secondly,it determined the weights of attributes according to their information entropy.Finally,all chosen attributes were integrated to determine whether the account pair was matched one.Theoretical analysis and experimental results show that,compared with subjective weight-directed objective weighting method based user identification algorithms,the proposed algorithm has better robustness and comprehensive evaluation metric when used to identify accounts between multiple social networks.3.An interests evolution rules analysis based user identification across multiple online social networks algorithm was proposed.Firstly,the proposed algorithm combined the activity categories with network structure as users' additional information in a social network to improve the standard topics mining model.Then it divided the user interests' topic distribution according to time period,and match accounts based on the combination of global static analysis and local dynamic analysis.Finally,experimental results on micro blog social networks show that the improved topics mining model in the proposed algorithm has lower perplexity value than the standard topics mining model,and when compared with the representative topic mining based user identification algorithms,the proposed algorithm has higher precision and better comprehensive evaluation metric.