A Hybrid Approach to Fake News Detection: Text and Images

Abstract

The widespread proliferation of fake news is considered one of the critical issues of the digital era, which affects public perception and decision-making processes. The presented work proposes an innovative approach to fake news detection, leveraging both textual and visual modalities with three state-of-the-art models: the Late Fusion Model, the Early Fusion Model, and VL-BERT. The models try to include both textual content and associated images of news articles to provide a complete system for spotting fake news with VL-BERT achieving the best overall performance. Hence, it focuses on the integration of visual and textual cues toward fake news detection, highlighting how multimodal methods can enhance the accuracy of the detection. The results are indicative that the multimodal incorporation of various data types yields a stronger solution toward mitigating this escalating issue of fake news in the online environment.