Machine Learning Models to Detect Online Fake News: a Systematic Literature Review

Abstract

Introduction: Including trusted and un-trusted information published by numerous sources, online content is disseminating more information than ever before. To counter this problem, we need to know the methods and techniques developed in the Detection of Fake News (FND). In this paper, machine learning (ML) models are examined to detect fake news and information. The impact of the fake news can be very severe. It has been used to jeopardize a Govt in past as well as create panic and may lead to financial as well as loss of human lives. Therefore, a comprehensive survey on the techniques to detect fake news has been performed in this research.
Objectives: We have provided a comprehensive summary of 15 articles based on a search of Pakistan's Higher Education Commission-approved journals from January 2019 to February 2022, which makes this SLR unique.
Methods: We examined the type of content, such as text or visual. Additionally, we identified several questions for research (RQs) the results of which will assist researchers in finding new challenges and gaps in fake news detection. Moreover, with this SLR, researchers will be able to gain a more comprehensive view of existing machine learning models and how they can be improved.
Results: There have been a number of techniques used for classification. SVM has been found to be the most dominant one. Python has been the most dominant language followed by other tools such as R and matlab for implementing research prototypes. A large variety of datasets and languages such as English, Urdu and Arabic have been targeted for detecting fake news. Features such as TF-IDF, bag of words have been used to detect fake news. Various metrics such as precision, recall, ROC curve, F1 scores have been measured to analyze different classifiers.
Conclusions: A systematic literature review of fake news detection has been performed in this study. Various research questions about the technique, tools, languages and metrics have been answered.