Research And Implementation Of Fact Verification Technology On Social Media

With the rapid development of social media platforms such as Twitter,fake news can spread rapidly on the Internet,affecting people’s lives and judgments.Fake news refers to false reports and rumors on social media,including completely false information and serious distortions of real events.However,due to the limitations of time and professional knowledge,it is difficult for ordinary people to identify fake news from the vast amount of information available on the Internet.Therefore,it is very important to develop automated auxiliary methods to detect fake news at the initial stage of information spreading.Despite the great success of deep learning method in fact-checking models,there are still many challenges as followed.Firstly,most models classify long texts,but most of source texts are short texts on social media,and the information is noisy.Secondly,Existing studies lack the learning of tweet spreading structure information on social platforms.Some studies need to collect rich user re-posts for each source.However,most users tend to simply re-share source tweets instead of sharing personal opinions while re-posting.Last but not least,the existing research is lack of interpretability,in fact,it can provide strong evidence for the detection results,whether from the text semantics,text dissemination or suspicious user detection.In order to solve the above challenges and problems,the main work of this paper is as follows:(1)A fact detection model based on dual attention mechanism integrated with session tree structure information is proposed.The session tree structure is used to model the tweets and their comments in the social platform,and all tweets are organized into a sequence structure according to the time order in which the tweets are published.Token-level self-attention mechanism and tweet-level self-attention mechanism are creatively introduced to learn text content and session tree structure,so as to use coattention mechanism to realize pairwise interaction between source tweet and all comments at the same time.And output the top three comment sentences which are most worth checking,to provide some explanation for the test results.(2)A fact detection model based on Graph Convolution Network(GCN)and dual coattention mechanism is proposed.The GCN learns the propagation representation of the source tweet.Then the the model extracts the user features from users’ social profiles and group interactions,which embeds the re-posting user characteristics into the spreading of the source tweet.Finally,the dual co-attention is used to investigate the specific types of forwarding users involved in the specific part of the source tweet,as well as the interaction between the re-posting users.The model output the most worth-checking users,which can provide evidence and explanation for the fact detection results.(3)Based on the main work of this paper,an online fact checking system is implemented.The system uses the model proposed in this paper,supporting users to input source tweets and comment messages,source tweets and re-posts for real-time fact detection.The system has data scrawler to get extra information.The system is user-friendly designed,supports multiple user-defined configuration,and has a convenient interactive interface,which meets the industry application requirements.