Rumor Detection on Twitter Using Features Extraction Method

Rumors can widely and rapidly diffuse in social networks than in offline platforms. This work was an attempt to address this challenge by introducing an approach for rumors detection, which learns from the Twitter dataset. In other words, this paper is going to suggest new features were extracted from Twitter datasets based on user behavior, the propagation features and the temporal features that representing the social context to reduce variation in features. Also, the proposed work involves adopting an ensemble classifier as a more appropriate approach and for achieving a better performance compared to existing individual-based classifiers. Moreover, the proposed method processes unbalanced data sets to reduce their impact on detection algorithms and improve the classification accuracy rate. Experimental results on the PHEME dataset showed that the new features made the classification process more effective and accurate compared to many related works of rumor classification that use the same data. Experimental results achieved 78.54% accuracy as an average of the events used.