Knowledge-Aware Multimodal Explainable Fake News Detection

This paper addresses the limited interpretability of current deep learning-based fake news identification methods, which often fail to incorporate background knowledge embedded in the news. It leverages commonsense knowledge extracted from the Baidu Baike commonsense knowledge graph as evidence to clarify the detection process. To effectively utilize the external knowledge implicit in the news, relevant commonsense is extracted from the ConceptNet knowledge graph, providing essential background knowledge for detection. We introduce the Commonsense-based Multimodal Fake News Detection Network (CMFNDN), designed to integrate evidence, commonsense, text, and image information for effective fake news detection.