A Transformer-based Infrastructure for Youtube Misinformation Detection

Abstract

This thesis addresses the growing concern of misinformation on social media platforms, particularly regarding COVID-19 vaccines. Focusing on YouTube, the study proposes a framework for identifying and filtering COVID-19 misinformation on the platform by collecting data and modeling misinformation detection. The methodology includes creating four YouTube accounts (from which data was collected), a custom Chrome extension for data collection, and using supervised learning techniques for video-related data labeling and classification. Incorporating Large Language Models (LLMs) and transformers enhances the accuracy of the misinformation detector. The study investigates the effectiveness of YouTube’s tools for identifying and filtering COVID-19 misinformation. The findings can facilitate the development of more effective strategies for promoting public health and safety. The methodology includes using state-of-the-art technologies and techniques like word-embeddings, transformers, natural language processing, and machine learning. Fusing these approaches provides a novel, more sophisticated, and accurate approach to detecting and filtering COVID-19 misinformation on YouTube, preventing its spread.