DSS: A hybrid deep model for fake news detection using propagation tree and stance network

Abstract

Nowadays, online social media play a significant role in news broadcasts due to their convenience, speed, and accessibility. Social media platforms leverage the rapid production of a large volume of information and cause the propagation of untrustworthy and fake news. Since fake news is engineered to deceive a wide range of readers deliberately, it is not easy to detect them merely based on the news content. Hence, more information, such as the social context, is needed. Moreover, to limit the impact of fake news on society, it is essential to detect them as early as possible. In this paper, we have developed an automated system “DSS” for the early detection of fake news wherein we leverage the propagation tree and the stance network simultaneously and dynamically.Our proposed model comprises three major components: Dynamic analysis, Static analysis, and Structural analysis. During dynamic analysis, a recurrent neural network is used to encode the evolution pattern of the propagation tree and the stance network over time. The static analysis uses a fully connected network to precisely represent the overall characteristics of the propagation tree and the stance network at the end of a detection deadline. The node2vec algorithm is used during structural analysis as a graph embedding model to encode the structure of the propagation tree and the stance network. Finally, the outputs of these components are aggregated to determine the veracity of the news articles. Our proposed model is evaluated on the FakeNewsNet repository, comprising two recent well-known datasets in the field, namely PolitiFact and GossipCop. Our results show encouraging performance, outperforming the state-of-the-art methods by 8.2% on the PolitiFact and 3% on the GossipCop datasets. Early detection of fake news is the merit of the proposed model. The DSS model provides outstanding accuracy in the early stages of spreading, as well as the later stages.