Automatic Rumour Detection Model on Social Media

Abstract

Social networking site Twitter, in particular, has become a popular spot for gossip. Rumors or false news spread very easily through the Twitter network by re-tweeting users without understanding the real truth. These reports trigger popular confusion, threaten the authority of the government and pose a major threat to social order. It is also a very necessary job to dispel theories as quickly as possible. In this research, multiple descriptive and consumer-based features via tweets are retrieved and integrated these features with the TF-IDF system to develop a composite set of features. This composite set of features is then used by several machine learning techniques like Support Vector Machine (SVM), Linear regression, K-Nearest Neighbor (KNN), Naive Bayes, Decision Tree, Random Forest, and Gradient Boosting. Along with these machine learning classification models, a Convolutional Neural Network (CNN) algorithm is proposed to distinguish rumour and non-rumor tweets. The proposed model is evaluated with freely accessible twitter datasets. The existing machine-based learning models have acquired an Fl-score of 0.46 to 0.76 for rumour detection, while the CNN model attained an Fl-score of 0.77 for rumour class. Overall, the CNN model yields greater results with a weighted average Fl-score of 0.84 for both rumour and non-rumor categories. The potential mechanism will help to detect misinformation as quickly as possible to counteract the dissemination of rumours and build users' deep confidence in social media sites.