Research On Expert Finding Method In Social Network Based On Topic And Context

Recently,online social network like Twitter and Sina Microblog influences people's daily life more and more closely.In such social network,everyone can post contents,and follows each other and affects each other as well.More and more people tend to ask the authority users in social network for advice when they have problems in daily life,and it makes expert finding problem have been widely studied.However,users usually want to find out experts on some certain topics like exercise and shopping when they retrieve experts in social network,while the relationship between users in social network is always related to various topics and mixed together.So these make it difficult to find experts on certain topics.Besides,most existing expert finding methods focus on social network which contains topic-irrelevant users and interactions.This results in that the expert results are not topic-specific and practical.Furthermore,contextual factors of social network,such as time and location and so on,also affect the accuracy of expert finding and are seldom concerned comprehensively in existing approaches.At last,the lifespan of experts and users' feedback activities toward expert results are seldom analyzed after expert finding in most existing studies on expert finding.To solve these three problems aforementioned,we propose a topic-specific contextual expert finding method,and analyze the lifespan of experts in expert finding results and users' feedback behavior.The main contributions and innovations of this paper include:Firstly,this paper constructs a topic-specific contextual feature model TSCFM which consists of a topic-aware model TAM for topical feature and a context-aware model CAM for contextual feature.TAM uses LDA and HITS to extract topical feature,and CAM evaluates social relation,time and location factors to extract contextual features.Secondly,based on TSCFM,we propose a topic-specific contextual expert finding method TSCEFM.An expert scoring function is learned by using SVM algorithm to synthetically concern both topical and contextual features.And we propose a rank method for the experts.The experiments on two datasets demonstrate that our proposed expert finding method is feasible and can improve the accuracy.Finally,the experts' lifespan and cascade feature prediction method is proposed based on the TSCEFM method.We analyze the underlying factors which affect the lifecycle of experts and users' feedback activities toward experts.An expert separability model is proposed to predict the lifespan of expert,and a user separability model is proposed for users' feedback activities prediction.The experiments prove the feasibility and accuracy of the proposed prediction method.