Authors: Dr. Prakash Kammam, Mukkapati Venu, K. Ashwini
Abstract: The fast spread of misinformation on social media platforms is causing serious issues with regards to public trust, democracy, and making sound decisions. The following paper provides a comprehensive overview of transformer-based systems for explainable misinformation detection, considering the latest research in the field of multimodal fusion, large language models implementation, and explainable AI. As can be seen from the systematic analysis, transformers surpass traditional methods in performance, with the multimodal system providing an accuracy of up to 94.5% and 81.1% on benchmark datasets. Moreover, large language models are very useful when generating background knowledge and enriching context, whereas explainability methods such as SHAP and LIME offer human-interpretable rationales for decisions made by the model. It was found that hierarchical progressive transformers successfully incorporate multimodality, combining different types of data such as text, images, background knowledge, and user comments, resulting in better performance than current methods.
DOI: http://doi.org/10.5281/zenodo.21338113
