Fake News Detection In Social Media Based On News Content And Users' Social Intercourse

The development of social media has greatly changed the way of information dissemination.In the traditional media era,users can only accept content passively.But in the social media era,users play the role of information receiver and creator,which makes social media have the characteristics of participation,openness and instant communication.With the above characteristics,social media accelerates the speed of information dissemination,expands the breadth and depth of dissemination,reduces the cost of dissemination,but also provides a suitable living environment for fake news.Considering the destructive effect of fake news on individuals and society under social media,fake news detection has become an important topic to be solved.This paper focuses on fake news detection in social media,selects Weibo dataset as the object,and studies from two aspects of news content and user social activities.In the part of dataset description and analysis,firstly,the information contained in each file in the dataset is summarized,the entities or concepts involved in the research are defined,and the news communication network is constructed based on the relevant entities;secondly,a network case is selected and visualized by Gaphi to show the structure and changes of the communication network;finally,the communication network is analyzed from the perspective of social network analysis.We calculate the statistical characteristics of network size and average path length of communication network.Through U-test,it is found that there are significant differences between the network size and average path length of fake news and real news.In the part of experiment,Word2vec + XGBoost and Text-CNN are used for the method based on news content,while Deepwalk + XGBoost,Node2vec +XGBoost and GAT are used for the method based on user social activities.A hybrid model is constructed by combining news content and user social activities.The results show that GAT and hybrid model have good performance,and the accuracy reach 85% and 90%.Secondly,in order to verify whether the GAT and hybrid model have the ability to identify fake news in the early stage,we build communication networks with different scale,and use the GAT and hybrid model to train and test the data.The results show that the two models have good detection ability when scale reaches 40%.Finally,in the experiment of exploring the importance of different types of features,it is found that the accuracy of using user related features is higher than that of using post related feature model.