Multi-modal Fake News Event Detection In Social Media

Social media has become an important tool for journalists to publish news,and it is also the main platform for people to obtain and share the latest news.However,the convenience and openness of social media have also led to the widespread dissemination of fake news,which has had a serious impact on society.With the improvement of network bandwidth and application technology,social media is flooded with more and more multimedia news.Compared with news that only contains text,news that contains multiple modal information such as text and images is more attractive.It is necessary to study how to use multi-modal information to detect fake news.Based on the multi-modal fake news data in real social media platforms,this paper proposes a related model for fake news event detection.This article mainly uses the multi-modal information and emotional characteristics of multiple posts in a news event to predict whether the corresponding news event is true or false.On real social media data sets,the model proposed in this paper has achieved remarkable results.The main work of this paper is as follows:(1)Construction of fake news event detection data set.This article obtained news posts from real social media platforms through web crawlers.In order to study the problem of multi-modal fake news event detection,the data set is preprocessed by deduplication and filtering,and finally the data is marked with event labels.The finally constructed data set used for multi-modal fake news event detection contains 2271 news events and 18005 online posts.(2)A multimodal fake news event detection network(MEDN)based on the attention mechanism is proposed.The MEDN model first uses deep neural networks to extract the text content features and visual content features of news posts.Then the attention mechanism is used to fuse these two features,that is,based on the correlation between text features and visual features,through visual features to better pay attention to the related text features,and finally merge into multimodal features.Finally,the multi-modal features of multiple posts in the news event are fed into the GRU network together as input,and a self-attention mechanism is introduced to better capture the features of multiple posts,and finally the multi-modal features of the news event will be learned to classify and judge whether the news event is true or false.(3)A multimodal fake news event detection network fused with emotion(emo-MEDN)is proposed.Emotional features play an important role in fake news detection tasks.This paper not only considers the multi-modal features of text content and visual content,but also introduces emotional features on this basis to further improve the performance of fake news event detection.This article integrates emotional features at two levels,post level and event level,and specifically compares the effects of the two fusion methods on performance improvement.The emo-MEDN model uses the emotional dictionary to extract emotional features in social media texts.There are two specific methods:a method based on emotional scores and a method based on word vector representation.(4)This paper compares the proposed MEDN model and emo-MEDN model with related baseline models on the constructed social media fake news event detection data set.The experiment proves that the MEDN model in this paper can effectively predict fake news events,and at the same time verifies the important role of the attention mechanism in the model.In addition,the experiments of the emo-MEDN model show that emotional features can improve performance,and the way to introduce emotional features at the event level performs better.