Fake News Detection On Social Media Based On Transfer Learning

With the rapid growth in the number of social media users on the Internet,the number of news generated on social media every day is also explosive growth.However,on the current social media,the number of users is too large,and the spontaneous spread of news is too fast,so the social media platform cannot verify the authenticity of the news one by one.This brings opportunities for the fake news to spread on the Internet.Although the current fake news detection models can curb the spread of fake news with obviously effect,but the current fake news detection modes still have following shortcomings: First,most of the current fake news detection models need a large amount of annotation data to train.When the annotation data is insufficient,the accuracy of the current fake news detection model is not high.Second,the self-learning ability of current fake news detection models is poor.It is difficult for the current fake news detection models to timely correct its own parameters according to the text characteristics of new events at the early stage of the event.So at the early stage of the event,the accuracy of the current fake news detection models is not high.This paper studies the above problems,and the main work of this paper is summarized as follows:(1)Aiming at the problem that the accuracy of the current fake news detection models is not high when the annotation data is insufficient,we introduce the transfer learning to the fake news detection model to improve the accuracy,and propose a fake news detection model BTUS,which based on model transfer.The BTUS cam obtain a strong text feature extraction ability by transferring the pre-trained BERT model.BTUS model combines the text features,user features and propagation features of news together as the auxiliary information for judging the authenticity of the news.(2)Aiming at the problem that the accuracy of the current fake news detection models is not high at the early stage of the event,we introduce the adversarial transfer learning to the fake news detection model to improve the self-learning ability of the models.We propose another fake news detection model BC-ACGAN,which based on adversarial transfer learning.BC-ACGAN model transfers the text features of the original annotation data,and make the text features distribution of the original annotation data tend to be close to the text features of the news belong to the new event.BC-ACGAN can timely correct its own parameters according to the text characteristics of new events at the early stage of the event.(3)The experimental results show that the BTUS model can improve the accuracy of the fake news detection model when the annotation data is insufficient.The experimental results also show that the BC-ACGAN model can improve the self-learning ability of the fake news detection models,so the accuracy of the fake news detection models can be improved at the early stage of the event.