A Comprehensive Approach to Automated Fact-Checking of Podcasts

Abstract

Podcasts have become a prominent medium for disseminating information, yet they remain largely uncharted territory for automated fact-checking systems. This research addresses the critical need for reliable fact-checking methodologies tailored specifically for podcast content. The study aims to enhance fact-checking accuracy and efficiency by developing high-quality datasets and leveraging existing Artificial Intelligence (AI) models. A comprehensive approach was taken to create datasets for claim detection and stance detection within podcasts. The evaluation of various AI tools revealed that, while existing models perform reasonably well on podcast data, there is significant room for improvement. Transformer models, when fine-tuned with podcast-specific datasets, showed enhanced performance in claim detection and stance detection tasks. The study also introduced a method for generating fact-check summaries for podcasts, enhancing the transparency and accessibility of fact-checked information. The results indicate that the accuracy and efficiency of podcast fact-checking can be significantly improved with more comprehensive datasets and advanced AI models.