Research On Fake News Detection Methods For Social Media

With the popularity and rapid development of the Internet,people’s access to information has gradually changed from traditional media such as newspapers and television to online social media.The low cost and timeliness of publishing and obtaining information on social media bring convenience to people’s lives,but at the same time,it inevitably promotes the emergence and dissemination of a large number of false information.A large number of fake news on the internet will not only guide public opinion,but even affect the stability of the country and society.Therefore,fake news detection for social media has important research significance.The current fake news detection methods mainly have the following shortcomings: On the one hand,many research methods only rely on monolingual evidence to detect fake news.However,news on social media usually involves different languages and cultural backgrounds,and insufficient information contained in monolingual evidence will limit the detection effect of the model.On the other hand,most researches on multimodal fake news only focus on the extraction and fusion of multimodal features,ignoring the effective use of external evidence.In view of the above background,this paper puts forward two fake news detection methods from the perspectives of multilingual and multimodal.The main research contents of this paper are as follows:(1)Aiming at the problem of insufficient evidence information in fake news detection based on monolingual evidence,this paper proposes a pre-training model based on multilingual evidence.First of all,this model connects news with evidence in different languages,and expands them as a whole through various data enhancement methods.Then,the feature representation of news-evidence pair is obtained by pre-training model,and finally the further representation of news-evidence pair is obtained by comparative learning framework to realize the classification of fake news.(2)Aiming at the problem that the existing multimodal fake news detection work ignores the effective use of external knowledge,this paper proposes a multimodal fake news detection model based on multi-task learning.The model is mainly divided into three parts: visual representation learning,text representation learning and multimodal feature fusion.Among them,in the text representation learning module,external evidence is used as a bridge between text and image,and the mutual enhancement between different modes in news posts is realized by modeling image-related evidence and news text.In addition,this paper attempts to use the multitask learning method for the first time to jointly model two tasks: fake news detection and evidence accuracy classification,in order to improve the performance of fake news detection tasks.The research scheme of fake news detection proposed in this paper makes full use of multilingual information and multimodal information,and takes contrastive learning and multitask learning as auxiliary methods,which effectively improves the detection effect of fake news detection tasks.