FramedTruth: A Frame-Based Model Utilising Large Language Models for Misinformation Detection

Guan Wang, Rebecca Frederick, Boshra Talebi Haghighi, B. L. William Wong, Verica Rupar, Weihua Li, Quan Bai

Abstract: The proliferation of misinformation jeopardises social cohesion, distorts the truth, and destabilises democratic processes. There have been numerous studies focused on detecting misinformation within online social networks. Nevertheless, much of this misinformation is portrayed from factual information but presented in a misleading manner, implying meanings that differ from literal interpretations and thus leading readers astray. Such instances of misinformation pose a significant challenge for classification since their textual features are very similar to those of truthful information. In this paper, we aim to address this challenge by proposing a deep-learning-based model, called FrameTruth, to detect misinformation originating from accurate facts portrayed under different frames. The proposed FrameTruth leverages Large Language Models to extract the framing of the information, incorporating this as an important feature in the process of misinformation classification. We evaluate FrameTruth by comparing its performance against a number of popular baselines using two real-world datasets. The experimental results demonstrate the superiority of our model in classifying the misinformation derived from the facts.