Social Media Rumor Detecting Base On Multimodal Deep Learning

With the popularization of mobile devices,the channels for obtaining hot news have changed from TV broadcasts and newspapers to mobile phones and tablets,and social platforms on mobile devices have also become the main channels for users to share and spread news.Due to the lack of effective supervision and the time gap between users to follow hot news.Criminals use this to create and spread rumors on social media.In recent years,spreading rumors on social media has been seen as a serious problem.Therefore,the main research of this article is to be able to identify information on social media in a timely and efficient manner.Text semantics is the main area of social media rumor detection.The development of deep learning has made rumor detection based on neural networks the mainstream.Most existing methods widely use recurrent neural networks(RNN),such as gated recurrent units(GRU)and long short-term memory networks(LSTM).A large number of recurrency leads to a significant decrease in the concurrent performance of the model,which means increased time and resource consumption.At the same time,the recurrent neural network cannot effectively extract the semantic features of the global text.This paper proposes a rumor detection model fused by multi-layer Transformer coding blocks.The self-attention mechanism in the Transformer coding block can provide better concurrency for the model.On this basis,the image features in the rumor events are learned through the Vision-Transformer,and the rumor detection method based on multi-modal deep learning is proposed by fusing the image features with the text features.This model can further improve the accuracy of rumor detection by using text information and image information of rumors.In addition,the propagation of social rumors on social platforms is based on social networks.This paper uses the graph neural network to study the features of the propagation of rumors,through the top-down,that is,the rumor information source toward the individual that spreads the rumor information,and the bottom-up,that is,the individual who spreads the rumor information learns from the two-way spread of the rumor.A rumor detection method based on graph Transformer neural network is proposed.Compared with the existing rumor detection models based on semantic features of text,features based on the fusion of multimodal and propagation features,the proposed model is superior to all the baseline methods,and the validity and rationality of the three models proposed by text are verified.The research in this article can find rumors accurately and timely,and can play a certain role in purifying the network environment.