Abstract
Social networking sites are a fertile ground for the fake news production and dissemination. Just like real world, differences about various issues are numerous among social media users at individual and group level. In order to strengthen position with respect to certain issue, an ugly approach is to fabricate fake news. The rapid spread of news about the event that never took place, can play negative role in very crucial occasions like elections where voters make their mind in the light of available information. The extent of fakeness can differ varying from slight twisting of information to one hundred degree opposite reporting of the event. Verification of credibility of information is a challenge that is being faced by internet users. One of the steps towards attempting to solve the challenge is to detect potential fake news as early as possible. In this paper, we used text mining techniques that have proven themselves to be efficient in many domains, in order to detect fake news by analyzing comments of people responding after reading the news on social media. The experiments were performed in two environments namely RapidMiner and Python to tackle the binary-class problem of assigning "fake" or "real" to a news. Arabic news dataset was especially constructed for the purpose of the research work. The NB classifier in RapidMiner environment was the most promising classifier achieving accuracy of 87.18% whereas SVM classifier outperformed other classifiers attaining 87.14% accuracy in Python environment. The output of the research work is expected to help Government authorities to automatically detect Arabic fake news on social media with less man-power.