User Credibility Evaluation Method Based on Factor Graph Model

Abstract:

In order to overcome the inability of traditional methods to effectively evaluate user credibility in the absence of user personal information and published content， a user credibility factor graph model （UCFGM） based on comment feedback and trust relationship is proposed， which formalized the influence of trust relationship and comment feedback on user credibility evaluation into a probability model， and proposed a semi-supervised classification learning method to build the model. The effectiveness of the model is verified on the Extended Epinions dataset， and it is found that the trust relationship is more likely to have a positive impact on user credibility evaluation than comment feedback. Compared with traditional algorithms， UCFGM can improve the accuracy of user credibility evaluation by 12% to 29% in the absence of user description and comment information.