Continual Graph Convolutional Network for Text Classification

Graph convolutional network (GCN) has been successfully applied to capture global non-consecutive and long-distance semantic information for text classification. However, while GCN-based methods have shown promising results in offline evaluations, they commonly follow a seen-token-seen-document paradigm by constructing fixed document-token graph cannot make inferences on new documents. It is challenge deploy them online systems infer steaming data. In this work, we present continual GCN model (ContGCN) generalize from observed documents unobserved Concretely, propose all-token-any-document dynamically update the every batch during both training testing phases of an system. Moreover, design occurrence memory module self-supervised contrastive learning objective ContGCN label-free manner. A 3-month A/B test Huawei public opinion analysis system shows achieves 8.86% performance gain compared with state-of-the-art methods. Offline experiments five datasets also show can improve inference quality. The source code will be released at https://github.com/Jyonn/ContGCN.