Text Classification Method Based on Hypergraph Convolutional Network and Attention Mechanism

Existing text classification methods based on graphs and hypergraphs have issues such as incomplete capture of text feature information, insufficient expression of multiple relationships, and inadequate attention to important information. Therefore, we propose a text classification model based on hypergraph convolutional networks and attention mechanisms, i.e., HCWA. Firstly, we construct three types of textual hyperedges for the text, including sequential hyperedges, thematic hyperedges, and syntactic hyperedges, to form multi-layered textual expressions and enrich the captured textual feature information. Secondly, we process the data using a hypergraph convolutional network to transmit and aggregate different relational information between nodes, enhancing the understanding of textual information. Then, attention is extracted from the nodes to improve the representation weight of important features, obtaining more critical features within the hypergraph. Finally, multi-channel pooling can capture overall features while preserving important features, reducing the loss of textual information. Extensive experimental results on real datasets verify that HCWA can effectively improve the performance of text classification.