| The rapid development of the Internet has broken the professional threshold of information communication,which makes more and more people acquire knowledge,share and express information through social media,greatly enriching people’s daily life.However,due to the huge number of users on social media,a lot of fake news fabricated for various purposes is made every day.What’s more,with the advancement of technology,fake news is no longer spread simply in the form of text but spread through a combination of text,pictures,and videos,which greatly enhances the confusing nature of fake news and makes fake news detection methods based on text content for analysis failed.Now,most fake news detection methods based on single modality view text as word sequences to extract features,and some multimodal methods simply concatenate the visual features and textual features of a post to get a multimodal feature and detect fake news.Most of them ignore the background knowledge hidden in the text content of the post which facilitates fake news detection.To address these issues,in this dissertation,based on the graph neural network,the following researches for multimodal fake news detection are carried out:(1)In the dissertation,we propose a novel Knowledge-driven Multimodal Graph Convolutional Network(KMGCN)to model the semantic representations by jointly modeling the textual information,knowledge concepts,and visual information into a unified framework for fake news detection.We convert text content in a post into a graph,which can model non-consecutive phrases for better obtaining the composition of semantics.Besides,we not only convert visual information as nodes of graphs but also retrieve external knowledge from real-world knowledge graphs as nodes of graphs to provide complementary semantic information to improve fake news detection.We utilize a well-designed graph convolutional network to extract the semantic representation of these graphs.The experimental results illustrate the validation of KMGCN.(2)Based on the KMGCN,we introduce graph attention and a pretrained BERT model.We propose a knowledge-driven adaptive multimodal graph convolutional network(KMGAN)for fake news detection.Compared with the predefined graph structure in KMGCN,KMGAN utilizes a graph attention mechanism that dynamically updates the weights between edges,which is beneficial to adaptively learn the topology of the graph and improve the flexibility of our model.The pretrained BERT model can better obtain contextual information for text features to improve the model.The experimental results illustrate the validation of KMGAN.To evaluate the proposed methods more objectively,we use several well-designed elimination experiments to show the effectiveness of components in Our model,and we show examples of the recognized fake news to demonstrate the effectiveness of the multimodal fake news detection in our model. |
