Abstract
Twitter is currently one of the most popularmicroblogging platforms allowing people to post short messages, news, thoughts, and so on. The Twitter user community is growing very fast. It has an average of 328 million active accounts today, making it one of the most common media for getting information during any influential or important event. Because it is freely used by the public, some credibility checking is required, especially when it comes to events of high importance. Automatic rumor detection in Arabic tweets is a challenging task due to the changes in the structural and morphological nature of the Arabic language, which makes the detection of rumors more difficult than in other languages. In this article, we proposed an effective approach for rumor detection of Arabic tweets using an eXtreme gradient boosting (XGBoost) classifier. We conducted a set of experiments on a public dataset that contained a large number of rumor and non-rumor tweets. The model uses a comprehensive set of features, including content-based, user-based, and topic-based features, allowing one to look at credibility from different angles. The experimental results demonstrated that the proposed XGBoost-based approach achieves 97.18% accuracy on 60% of the dataset as a training set, which is the highest accuracy rate compared with the other methods used in recent related work.