Research on Text Classification Based on Improved Graph Convolution Neural Network

Compared to traditional text classification methods, Graph Convolutional Neural Networks (GCNs) have an advantage in handling graph-structured data, making them more effective at capturing the deep semantics and structural information of long and complex structured texts. However, they still face challenges in effectively constructing text graphs and utilizing local structural information in practical applications. To address this challenge, this paper proposes an improved Graph Convolutional Neural Network (GCN) model—PT-InducTGCN (PPMI-TF-IWF-Inductive Graph Convolutional Networks) model. The model optimizes the shortcomings of traditional methods in capturing textual semantic relationships by constructing heterogeneous graphs. The PT-InducTGCN model uses the Term Frequency-Inverse Word Frequency (TF-IWF) algorithm instead of the traditional TF-IDF to more accurately measure the relationship between words and documents, and enhances the analysis of word co-occurrence relationships through the Positive Pointwise Mutual Information (PPMI) algorithm. By integrating the new document representation and heterogeneous graph construction into the GCN framework, experiments show that the model not only improves classification accuracy but also demonstrates effectiveness in inductive text classification tasks on various benchmark datasets such as R8, R52, Ohsumed, and 20NG.