Research Of Fake News Detection Based On Short Text Data Document Representation Generation

In the Internet era,every netizen can release news through social media and other channels.The ideology and behavior of grabbing news,grabbing exclusivity,lead to an endless stream of fake news,which will not only mislead readers who do not know the truth,but also cause bad social impact.Effective control of fake news has become an important proposition in creating a clear cyberspace.The intelligent detection method has the advantages of fast speed and high recognition accuracy,which has been widely used in fake news detection.In the fake news detection,the intelligent detection model focuses on the extraction of text features and the use of text information.Current researches mainly have the three problems:First,there are a large number of short text news in social media,and these short news have the weak description information,resulting in sparse text features,which leads to the model have difficulty in extracting text features from short news;Second,the existing models only focus on the local information of the word when the text is vectorized,and lack of long-distance and discontinuous text word interaction,resulting in the discontinuity of text semantic information;Last,When judging the authenticity of news,the model only rely on the content description of news and ignore the auxiliary role of external knowledge in fake news detection,which results in low recognition accuracy when the model detects news in specific fields.To solve the above problems,this thesis has carried out the research of fake news detection based on short text data document representation generation.The main contributions of this thesis are as follows:(1)Intra-graph and Inter-graph Joint Information Propagation Network with Third-order Text Graph Tensor for Fake News Detection(abbreviated as IIJIPN).This method first proposes a third-order text graph tensor,which extracts sequential features,syntactic features and semantic features from the text information,solving the problem of sparse text features.Intragraph information propagation is firstly performed in each graph to realize homogeneous information interaction by building graph for each text property;Secondly,inter-graph information propagation is then performed among text graphs to realize heterogeneous information interaction.By performing two kinds of information interaction,the model can obtain continuous context semantics in the news,and solve the problem of lacking long-distance text interaction.(2)Hierarchical Text Representation Generation Network with external knowledge for fake news detection(abbreviated as HTRGT).This method effectively solves the problem of lack of external knowledge reference by referencing the description of news entities in external knowledge bases such as Wikipedia or Oxford Dictionary and comparing it with the contextbased entity description to judge the difference of entity semantics and generate entity representation.In addition,the method also focuses on the document structure.Based on the context-based entity representation,the hierarchical encoding mechanism generates the entitydocument vector representation,and constructs the continuous semantics in the text.In order to evaluate the performance of the model,multiple sets of comparative experiments were conducted on four public fake news datasets.The experimental results show that the IIJIPN proposed in this thesis solves the problems of short text feature sparsity and long-distance text word interaction deficiency,and HTRGT solves the problems of external knowledge reference deficiency and long-distance text word interaction deficiency,effectively improving the accuracy of fake news detection.