Research On User Identification Algorithm Across Social Networks

Becoming an indispensable part of our lives gradually,various social networks have sprung up,in recent years,such as Weibo,Douban,Zhihu and so on.As an important part of study on online social networks,algorithms of user identification are of vital importance in many study areas,such as cyberspace security,personalized service recommendation and data mining on social network.At present,user identification algorithms on cross-social network have made a lot progress.Various user identification algorithms have been proposed.While there are still some deficiencies in these studies,such as the inconsistency of tags,the instability of matching and the one-sidedness on extraction of profile feature.In this dissertation,two user identification algorithms are proposed to solve problem bring by the inconsistency of tags,the instability of matching and the one-sidedness on extraction of profile feature.Firstly,in order to solve the inconsistency of tags,the user identification algorithm based on semantic similarity of user tags is proposed.In the first place,the algorithm uses similarity of the user name as the assessment indicator to select candidate matching accounts;In the second place,similarity feature between tags of user are calculated based on semantic similarity algorithm of tags.In addition,the topic generation model is used to extract topic keywords from the content published by user as tags of user when tags are lost;In the third place,the algorithm determines the matching of two accounts by calculating whether the tag similarity of the two accounts is greater than a certain threshold.Compared with the traditional tag-based user identification al?gorithms,the mean reciprocal rank is increased by 15%,and the accuracy,precision,recall and Fl-score are increased by 22%,15.7%,24%and 20.5%,respectively.Secondly,in order to solve the instability of matching and the one-sidedness on extraction of profile feature,the user identification algorithm based on stable matching problem is proposed.In the first place,candidate accounts are selected by using a combination of the username and the network;In the second place,multiple similarity features are extracted from user profile,including string,semantic and different attributes similarity;In the third place,the two-sided matching algorithm is used to achieve account matching.Compared with the related algorithms proposed by the predecessors,the precision,recall and F1-score have increased by 2%,28%and 18.8%,respectively.Finally,we draw conclusions from the discussions and give an expectation of this field.