Cascade-LSTM: A Tree-Structured Neural Classifier for Detecting Misinformation Cascades-KDD2020

　　Cascade-LSTM It is a tree structure neural classifier for false information cascade detection , It is essentially a rumor （ Fake news ） Test model , It regards the rumor detection task as a tree classification problem .

 　　Cascade-LSTM In recurrent neural networks （ This article is based on TreeLSTM, Tree structured LSTM） On the basis of , Introduced a two-way TreeLSTM Structure to code the user characteristics of the propagation tree nodes by traversing from bottom to top and from top to bottom along the propagation tree structure .

        In particular ,Cascade-LSTM First, traverse from leaf node to root node from bottom to top , Update node characteristics , Then traverse from root node to leaf node - Take the hidden state vector from bottom to top 、 The node feature and the implicit state vector of the parent node are the inputs , Update the node characteristics again , To encode the context dependency between nodes , To capture the propagation dynamics .

        here ,Cascade-LSTM take The hidden state vector converging to the root node from bottom to top and The average of all root nodes traversed from top to bottom Implicit state vector as output . Besides ,Cascade-LSTM Added source text features （ Emotional characteristics of text ）.

        It is worth mentioning that ,TreeLSTM In essence, it can be regarded as GNN, and Cascade-LSTM Also similar , But their message aggregation is directional .