Rumor Detection Based on Depth and Breadth with Tree-Structured Recursive Neural Networks

With the rapid development of social media, it is very important for automatic rumor detection to obtain global information about rumor propagation. However, to get global information on propagation, it is necessary to use the propagation structure rational. In this paper, we propose a novel rumor detection model, named depth-breadth tree-structured recursive neural networks(RvNN-DB). the rumor propagation tree can be used to extract the depth propagation features of multi-branch and the breadth features at the same depth, effectively obtaining the global information of propagation; In extracting the depth propagation features, we introduce attention mechanism to adjust the weight of clues so that the model can focus on the key clues that are conducive to rumor detection; In extracting the breadth propagation features, one-dimensional convolution neural network is used to obtain the fine-grained relationships among word levels within the comments at first, obtain the semantic interaction between the comments in the next step, the purpose is to supplement the depth propagation features. Results on two public Twitter datasets demonstrate that our method has higher detection accuracy in early rumor detection than the existing linear propagation structure and unidirectional rumor detection network in the propagation tree.