Fake News Detection Based On The Publisher-user-news Ternary Relationship Network

Social media has become a new base for the spread of fake news.Compared with traditional channels of news communication,social media spreads faster,has a wider scope and has a greater influence.The widespread spread of fake news not only undermines the credibility of the news media and the government,but also has an impact on social security and public life.Therefore,how to automatically identify fake news on social platforms has become an important but challenging problem for the industry and academia.At present,there are three recognized in the detection problem of false news features:the news content,users and publishers,but most of the existing detection methods only focus on the one from the three aspects of the features of customized solutions,not make full use of the correlation among information,thus to some extent,restrict the effect of the model and versatility.Starting from the social background in the process of news dissemination,this paper will combine the news,users and publishers,and dig into the potential relationship between user-news and publisher-news.In the process of research,we found that existing models tend to use machine learning methods and deep learning approach,rarely from the perspective of mathematical model,only Shu^[1],etc.Tri?FN model put forward in 2019,not only the integrated use of the above three characteristics,and the relationship between them using embodied the mathematical model in this paper.However,the application of this model is based on English,and the selection of some features is not applicable to China's national conditions.Moreover,the model on the relationship between publishers and news only utilizes a static background attribute of publishers,ignoring that publishers themselves are also users of social media,and there will be social relations.This paper makes some improvements on the basis of Tri?FN and proposes a network-based ternary relational embedding model Tri?NET.Based on the original model,this model considers the characteristics of publishers comprehensively.On the one hand,it uses the attributes of publishers to calculate the reputation score of publishers,and on the other hand,it introduces the following relationship between users and publishers,so as to realize the comprehensive mining of the role of publishers in news classification from the two aspects of background and social interaction.Secondly,in order to more accurately obtain the social relationships formed between users based on news,this paper uses the social relationship theory combined with the method of similarity analysis to build a social network.In addition,the selection of features comprehensively considers the situation of China's social media,and replaces some features that are not applicable to China's national conditions in the original model.Finally,experiments on the microblog dataset prove that the improved model in this paper is superior to the original model and significantly superior to other baseline methods.