Chinese Text Classification Method Based on Graph Embedding Residual Bidirectional Long Short-Term Neural Network

In response to the problem that graph neural networks often focus more on the relationships between local nodes in Chinese text classification tasks, while the relationships between distant nodes may be overlooked, this paper proposes a Graph Embedding Residual Bidirectional Long Short Term Memory Network (GRBiLSTM), a fusion model suitable for Chinese text classification. This model adopts a dual embedding strategy for feature construction of Chinese text, and also refers to the concept of residual networks to optimize the BiLSTM model, aiming to alleviate the problem of early feature forgetting that may occur as the number of layers increases during the model training process. By conducting experiments on multiple public Chinese text classification standard datasets and comparing with multiple mainstream text classification methods, our method has shown better experimental results. Multiple ablation comparison experiments have also been conducted to further verify the effectiveness of our method.