Utilizing Continual Learning with Unlabeled News Feeds: A Case Study on Rumour Detection

Abstract

Rumours are everywhere, and their detection is not an easy task at all. There are both manual rumour detection approaches like reader-added context supported by some social media platforms, and automatic approaches like AI models that detects rumours using various architectures. Both methods take some time to detect and label a piece of information as a rumour. In this paper, we present a novel approach for realtime rumour detection in news feeds using a combination of RoBERTa and Elastic Weight Consolidation (EWC), a continual learning algorithm. Our method integrates data collected from news feeds via the ‘news-please’ scraper and augments it using LLaMa 3, a Large Language Model based data augmentation technique. This framework can help in solving numerous problems and as a case study we use it to provide a robust solution for addressing the challenges posed by misinformation propagation in the digital age, offering a promising tool for news verification and trustworthiness assessment in online information ecosystems.