Research On Multi-modal Fake News Detection Based On Knowledge Graph

With the rapid popularity of the Internet,it is more convenient for people to get news from online social media than other traditional media.However,in the absence of effective supervision and supervision,the open Internet promotes the spread of a large number of fake news.Fake news in social media spreads rapidly in the Internet at a very low cost,which will bring significant negative impact on society and people's daily life.Therefore,it is of great practical significance to design an automatic detection model for fake news.On the one hand,the traditional single-modal fake news detection models only focus on the statistical and linguistic features of the news text itself,and detect false news by machine learning or deep learning modeling based on the features mined.On the other hand,these models mainly use image statistics and image distributed representation features,There is no deep mining of the meaning of words and knowledge behind the image.In view of the shortcomings of the existing methods,the following work is carried out in this paper(1)This paper proposes a fake news detection model based on knowledge graph.The model supports the false news detection task by introducing external knowledge base,generates vector representation for the extracted news text triples,and uses cosine to calculate the similarity between triples.Finally,the similarity information is used to detect false news.(2)This paper proposes a multi-modal fake news detection model based on deep learning framework which combines knowledge graph and image caption.The model not only extracts the triple-style knowledge graph in news text,but also integrates the description text generated by image.At the same time,the original text,triple and image description text are integrated by using Bert framework.The experimental results on benchmark Chinese false news corpus show that the model is significantly outperforms the existing representative methods.