Multimodal Fake News Detection Technology Based On Deep Learning

With the development of science and technology,incidents of fake news triggering social crises occur from time to time.The change of media media makes false news gradually change from text form to multi-modal form with graphic and text coexisting.Multimodal fake news carries heterogeneous forms of text and image information,and research on the multimodal content of fake news can improve the effect of fake news detection.Existing multimodal fake news detection methods have the following deficiencies.First of all,in terms of feature extraction,most methods do not consider their own effective features,and features are not universal at this stage.Secondly,in terms of feature fusion,in the previous fake news detection methods that used image features combined with text features,the feature information in the news was simply spliced,without considering the interactive information between the two modalities.Finally,in terms of model structure,few models can verify the effectiveness of extracted features.According to the problems encountered in the current research,this paper conducts multi-modal fake news detection research based on deep learning,which further improves the accuracy of false news detection.The main contents are as follows:(1)In terms of feature extraction,this paper analyzes the content characteristics of fake news on the basis of in-depth research on multi-modal feature extraction technology,and proposes a multi-modal feature extraction algorithm based on a dual-branch adversarial network.The algorithm extracts the feature information of the modality from different levels,and joins the domain confrontation network to improve the universality of the feature.(2)In terms of feature fusion,this paper proposes a multi-modal feature fusion algorithm based on a combined fusion mechanism on the basis of studying multi-modal feature fusion technology.The algorithm uses the multi-mode bilinear pooling method to perform information interaction between modalities to supplement the richness of features,and uses the self-attention mechanism to enhance the internal information of the modalities to improve the effectiveness of its own features and realize heterogeneous modal feature information.interaction and enhancement.(3)In terms of model structure,in order to improve the interpretability of the model,this paper proposes a fake news detection algorithm based on variational autoencoder for multi-task learning.A variational autoencoder is added to reconstruct the features,and then the loss function is improved and optimized by using the characteristics of multi-task learning,so that the model has generalization.(4)This paper conducts a large number of experiments based on two authoritative data sets,Weibo data set and Twitter data set,to verify the effectiveness of the proposed model and algorithm,and design a false news detection system for auxiliary detection.