Research On User Identification Of Social Media

Along with the development of mobile network technology,the user number of social media has seen explosive growth and social media has become a gigantic database storing user's information.User identification of social media refers to identify user's related contents like user's account,attribute,psychology state and behavior custom.With different social purposes,the same user usually registers multiple accounts on different social applications,and probably provides different identity information.Matching the same user's accounts from different platforms through user identification technique of social media,then constructing full and well-rounded user information library,can provide effective support for downstream applications like commodity recommendation,information retrieval and cyberspace supervision.Meanwhile,considering the protection of personal privacy,user usually tends to conceal his personal information,which leads to difficulty in directly acquiring some valuable attribute information of user.Conjecturing user's missing attribute information through user identification technique of social media enables further completion of user information library,and thus know more comprehensively about user.This thesis,focusing on user identification of social media,conducts research from two aspects,namely matching user accounts registered for multiple social media and user attribute identification of social media.The main contribution includes the following two respects:(1)A method of account matching based on Viterbi algorithm is proposed.Traditional accounts matching methods suffer from low matching efficiency and low matching accuracy for accounts from multiple platforms.This thesis solves problems existing in traditional methods through the following three steps.First,it constructs candidate matching accounts set based on username similarity of accounts,and solve the low matching efficiency problem of traditional method through narrowing down candidate matching accounts set.Second,it calculates similarity between accounts and constructs matching network based on registration information of accounts.Finally,it finds the most matching accounts in matching network through Viterbi algorithm,and realizes accounts matching from multiple platforms.Compared with traditional methods,matching accuracy for multiple platforms in this thesis improves to some extent.(2)A method of user attribute recognition based on graph embedding is proposed.Traditional user attribute identification methods generally adopt measures with supervised learning,and their attribute identification accuracy are relatively low and severely depend on number and quality of tagged data.This thesis solves the existing problems in traditional methods through the following three steps.First,it constructs heterogeneous network based on text theme and Wikipedia entity,and enrich the semantic information of heterogeneous network through introducing different types of nodes to heterogeneous network.Second,it acquires the embedding representation of user node in heterogeneous network through graph embedding algorithm.Finally,it conducts user attribute identification of social media based on embedding vector of user node.The method in this thesis belongs to semi-supervised learning,and its attribute identification accuracy is affected by tagged data to a relatively small degree.Compared with traditional methods,it acquires relatively high accuracy with relatively little training data,and solves the problem of attribute identification accuracy severely depending on tagged data in traditional methods.