RumorDetect: Detection of Rumors in Twitter Using Convolutional Deep Tweet Learning Approach

Abstract

Nowadays social media is a common platform to exchange ideas, news, and opinions as the usage of social media sites is increasing exponentially. Twitter is one such micro-blogging site and most of the early update tweets are unverified at the time of posting leading to rumors. The spread of rumors in certain situations make the people panic. Therefore, early detection of rumors in Twitter is needed and recently deep learning approaches have been used for rumor detection. The lacuna in the existing rumor detection systems is the curse of dimensionality problem in the extracted features of Twitter tweets which leads to high detection time. In this paper, the issue of dimensionality is addressed and a solution is proposed to overcome the same. The detection time could be reduced if the relevant features are only considered for rumor detection. This is captured by the proposed approach which extracts the features based on tweet, reduces the dimension of tweet features using convolutional neural network, and learns using fully connected deep network. Experiments were conducted on events in Twitter PHEME dataset and it is evident that the proposed convolutional deep tweet learning approach yields promising results with less detection time compared to the conventional deep learning approach.