Rumor Detection with Hierarchical Recurrent Convolutional Neural Network

Automatic rumor detection for events on online social media has attracted considerable attention in recent years. Usually, the events on social media are divided into several time segments, and for each segment, corresponding text will be converted as vectors for various neural network models to detect rumors. During this process, however, only sentence-level embedding has been considered, while the contextual information at the word level has been largely ignored. To address that issue, in this paper, we propose a novel rumor detection method based on a hierarchical recurrent convolutional neural network, which integrates contextual information for rumor detection. Specifically, with dividing events on social media into time segments, recurrent convolution neural network is adapted to learn the contextual representation information. Along this line, a bidirectional GRU network with attention mechanism is integrated to learn the time period information via combining event feature vectors. Experiments on real-world data sets validate that our solution could outperform several state-of-the-art methods.