Fake News Detection System In Social Networks

Social media is booming,and its practical,convenient,open and inclusive features have attracted a huge user group,making it an indispensable part of people's daily lives.However,openness and tolerance is a double-edged sword.While encouraging users to truly express themselves through social platforms,it has also become a cradle for criminals to spread false information.Due to the large user base of social platforms and the lack of effective management and control,information is easy to generate and spread quickly.And because social media is the main channel for people to obtain information,if false information is allowed to spread,the existence of the echo chamber effect will gradually distort the message in the process of dissemination.This will seriously disrupt people's understanding of the facts,and even affect politics and economy etc.This thesis collectively refers to assertive stories or statements on social platforms as news,and detecting fake news from Twitter is the focus of this thesis.Detecting fake news is essentially to assess the credibility of news.Existing fake news detection methods have proposed many effective methods from the perspectives of text content,network structure,and user characteristics.Based on the previous experience,the main work and innovations of this thesis are as follows:1.Taking the social platform Twitter as an example,we have conducted an in-depth study of its data representation and data collection methods,and proposed a method of dynamically constructing a tweet propagation network based on users' interaction networks and friendship networks,and finally using it to build high quality datasets which contains tweets,users and networks.2.The news dissemination network is decomposed based on the sparse representation method,and the difference between the atoms in the true and false news networks is found.Then the atomic features that can be used for fake news detection are extracted.Finally,based on the BERT model,the transfer learning method is used to classify real and fake news.The experimental results show that the fusion features we used improve the accuracy of classification compared to only using text content.3.In the process of research,we have also developed a system that integrates data collection,data analysis,visualization and other functions.On the one hand,it assists the progress of the research work of this paper,and on the other hand,it also lays the foundation for continuous research on the changing characteristics of fake news.Finally,our system has high versatility.