Explainable Evidence-Based Veracity Assessment of Textual Claim

Abstract

The rise of social media and the internet has significantly increased the amount and speed of shared information, posing challenges for verifying content. Automated veracity checking has become essential in quickly and accurately evaluating claims due to the overwhelming volume of data. The reliability of these systems depends on their ability to access and evaluate substantial evidence, which is crucial for authenticating assertions and preventing the spread of misinformation. This study proposes a new method that integrates rationales from evidentiary texts to address the issue of insufficient evidence in automated veracity checking. By using contextual coherence and relevance as metrics when direct evidence is limited, our technique aims to assess evidence sufficiency comprehensively. Furthermore, it goes beyond identifying evidence sufficiency by examining supporting or refuting rationales, enhancing our understanding of claim veracity. Our research introduces a preservation technique focused on maintaining contextual consistency and logical validity to overcome limitations in existing veracity-checking systems. This approach prioritizes alignment between claims and their evidence, effectively addressing issues related to evidence insufficiency by capturing subtle semantic connections while assessing contextually implied meanings often overlooked in traditional methods of evidence evaluation