Combating Misinformation in Social Media and News: A Deep Learning Approach

Abstract

Combating misinformation has become a critical challenge in today's information-driven society, particularly with the proliferation of fake news, propaganda, and biased content across various domains. This study explores advanced natural language processing (NLP) techniques, including feature extraction and selection, to analyze and classify datasets such as Q-Prop, ISOT, GRAFN, and PubHealth. The relief algorithm is employed for feature selection to identify the most relevant attributes, enhancing the efficiency of machine learning models. XLNet, a powerful transformer-based model, is utilized for document representation and classification due to its ability to capture bidirectional and long-term contextual dependencies. The proposed methodology demonstrates how robust embeddings, combined with domain-specific datasets and optimized feature selection, can accurately classify content across news, politics, and public health domains.