Sentiment Analysis Model for Fake News Identification in Arabic Tweets

Abstract

Over the last few years, the exponential rise of social media, particularly Twitter, is becoming a major source of news sharing and consumption among users. These platforms allow users to publish, author and distribute content. These environments may be used to report and spread gossip and false news, whether accidentally or maliciously. Fake news and inaccurate machine-generated text are serious issues affecting societies worldwide, particularly the Arab world. This motivates efforts to identify fake and distorted news. This paper aims to introduce a robust prediction model to identify fake news in Arabic Tweets. Several Natural Language Processing (NLP), feature selection, and advanced ML algorithms were exploited to achieve this purpose. NLP techniques were used to process and transform the given tweets into structured form. The recursive feature elimination (RFE) technique was employed to eliminate uninformative features. ML methods were used to build the prediction model. Experimental results revealed the superiority of the Logistic Regression (LR) classifier among other algorithms. Moreover, RFE proved its ability to enhance the overall performance of the LR classifier. Overall, the proposed model provided an acceptable prediction accuracy of 82%.