Event-Based Rumor Detection using LSTM Models For Arabic Content on Twitter

Due to the flexibility of sharing and exchanging information, Microblogging networks such as Twitter have become more and more popular over the last few years. The nature of social media as well provides a rich and facilitated ground for rumormongers to spread false stories and information that may result in major mess and unpredictable reactions from involved individuals. Detecting rumors automatically at their early formative and diffusion phases is particularly crucial to minimize their negative and catastrophic effects. Many effective solutions to detect rumors on Twitter have been presented over the last few years. However, most of the researches were limited to certain languages, such as English and Chinese. Few studies focused on the Arabic Language; Most of those studies were concerned only to detect rumors at the tweet-level and neglecting event-level based detection. In this paper, we built two RNN-LSTM deep learning models to classify Arabic Twitter content as rumor or non-rumor more efficiently utilizing linguistic features. The constructed models are capable to detect rumor events on Twitter. An attention mechanism was embedded to optimize the performance of these models where they achieved accuracy above 93%.