A Comprehensive Survey on Fake News Detection Using Machine Learning

Abstract

In the age of information data surplus and social media impact, the propagation of fake news has developed a significant societal concern. This challenge requires robust tools and methodologies for addressing, with machine learning emerging as a promising approach. The paper reviews different machine learning techniques in fake news detection, including supervised, unsupervised and semi-supervised methods. Supervised methods utilize labelled datasets to train models to discriminate between fake and legitimate news articles. Unsupervised Learning methods, on the other hand, rely on clustering as well as anomaly detection to identify suspicious patterns in information data. Semi-supervised techniques associate elements of both supervised and unsupervised learning techniques for leveraging limited labelled data successfully. Moreover, the manuscript examines feature extraction techniques, including Natural Language Processing (NLP) techniques like bag-of-words, word embedding and syntactic parsing. It also discusses the importance of incorporating contextual information, such as source credibility and social network dynamics, into the detection process. The paper addresses the evaluation metrics normally used to evaluate the performance of fake news detection models, such as accuracy percentage, precision, recall and F1-score. It highlights the need for robust evaluation frameworks to ensure the reliability and generalizability of the fake news detection system.