Rumour Detection on Twitter Using Long Short-Term Memory

Abstract

Microblogging platforms are perfect place for breeding and spreading rumours. Propagation of these rumours at an alarming rate has become a severe social issue which has an adverse effect on people and organizations. Therefore, it becomes important to detect rumours quickly, automatically and increase the accuracy of learning model than the existing models which uses feature engineering that is biased, labour intensive and time-consuming. Even though, recent studies uses machine learning-based methods for automatic rumour detection by extracting features of rumour contents (e.g., people’s opinions, questions, etc.) and static spreading processes, early detection of rumour remains a challenge. This paper proposes a learning model, Long Short-Term Memory (LSTM) combined with pooling operation of Convolutional Neural Network (CNN) for early detection of rumour. LSTM networks are well-suited for processing, classifying and making predictions based on time series data and also deals with vanishing and exploding gradient problems that can be encountered when training traditional Recurrent Neural Network (RNN). The dynamic changes of forwarding contents and spreaders are taken into consideration using LSTM based model.