Abstract:
The spread of misinformation on social media platforms endangers public discourse and societal trust. Addressing this issue requires robust mechanisms for identifying and verifying the accuracy of information circulated online. This thesis focuses on improving fact-checkers’ capabilities by creating a comprehensive framework for efficient misinformation management via identifying, simplifying, and verifying claims on social media. This research endeavour is multifaceted and involves (a) comprehending the nature and context of claims made on social media, (b) creating algorithms to detect these claims automatically, (c) simplifying complex or noisy claims to enable more accessible analysis, (d) determining which claims are worthy of verification, and (e) using sophisticated computational techniques to verify the accuracy of these claims. We employ natural language processing and machine learning algorithms to parse and interpret textual content from social media platforms. Using cutting-edge models, we automatically detect and distil claims from massive amounts of data, addressing the scale challenge inherent in social media ecosystems. Furthermore, we use novel claim simplification processes to convert verbose or ambiguous statements into explicit, concise claims that can be verified. In the final stage, we verify identified claims, thus providing a reliable method to affirm or refute claims. The efficacy of our methods is demonstrated through extensive experiments on real-world data, showing significant improvements in the speed and accuracy of fact-checking operations. This thesis contributes to fact-checking by offering a scalable, efficient solution for combating misinformation on social media and equipping fact-checkers with tools critical in the fight against information distortion. The methodologies presented herein lay the groundwork for future research and practical applications in digital information verification.
