An advanced ensemble approach for detecting fake news

Abstract

The explosive growth in fake news has evolved into a major threat to society, public trust, democracy and justice. The easy dissemination and sharing of information online provide the unabated momentum. As such, it has become crucial to combat the menace of fake news and to mitigate its consequences. Detecting fake news is an intricate problem since it can appear in a multitude of forms, thus making it both automatically and manually very challenging to successfully recognise. Furthermore, fake news is intentionally created to mislead and is often interspersed with real news. Studies have shown that human beings are somewhat unsuccessful in identifying deception. The majority of people accept that information they are presented with in virtually any form is reliable or veracious. The relevant literature reveals that a considerable number of people who read fake news stories report that they find them more believable than the news that is disseminated via mainstream media. Furthermore, there are predictions that by 2022, the greater population within mature economies are likely to consume more false than true information. The importance of combatting fake news has been starkly demonstrated during the current Covid19 crisis. Social media networks are significantly increasing their efforts to develop fake news detection mechanisms, as well as to enlighten subscribers on how to recognise fake news, however most people are naturally predisposed to spreading sensationalist news without any fact-checking process in place. It is therefore evident that the creation of automated solutions is vital and urgent for the detection of untruthful news and as such, the goal of this study is to aid in the detection of fake news. Prior studies have included many machine learning models with varying degrees of success but many non-conventional machine learning models have not yet been exploited despite evidence to suggest that they are the best in several text classification scenarios. Consequently, an ensemble learning approach is suggested to assist in resolving the gap that has been identified.

Contemporary studies are validating the efficiency of ensemble learning methods and have provided encouraging outcomes. This study investigates how machine learning and natural language processing methods are pooled together in a blended ensemble in order to build a model that will utilise data from past news articles, to forecast whether a current news article is likely to be false or true. A variety of performance metrics such as roc, roc auc, recall, precision, f1-score and accuracy are used in comparing the proposed model to other machine learning models. The measurements are applied in evaluating and gauging the efficiency of the proposed model. The results obtained show that the proposed model’s performance is better than several other learning models, which is very encouraging.