An Ensembled Text-based Multi-Stacked-CNN-BILSTM for Rumor Detection on Social Media

Social networks heave became a part of people’s life. Although there are many positive sides like eCommerce and easy communication through social networks, the existence of rumors is the negative side of social networks. With the least effort, a rumor can be quick to many users disseminated leading to unpleasant social phenomena. In the literature, there are some machine learning and deep learning approaches to automatically detect rumors. Most of the previous works fail to detect the rumors in high performance. In this paper, we present an ensembled multi-stacked-CNN-BILSTM model to detect rumors on social media. This deep learning model just considers the text of users and detects that whether a text message is a rumor or not. In the proposed model, each multi-stacked-CNN-BILSTM makes a prediction, and the label that voted maximum as predicted label is detected. We have evaluated the proposed model on the real public data set of PHEME. The proposed model achieved accuracy improvement over the ten state-of-the-art machine learning and deep learning models.