NE ZHA-TextCNN Method for Multi-Label Long Text Classification
Yuchen Ye, Xiaodong Chang, Amin Fan · Procedia Computer Science · 2025
To solve the problem of multi-label long text classification, a neural context representation- text convolutional neural network model for Chinese understanding is proposed. In order to comprehensively capture and represent the context information and semantic features in long texts, the model adds a global context level semantic information extraction module. In addition, the semantic information vector matrix is convolved by convolution kernels of different sizes to extract local semantic feature vectors, and the 1-MaxPoll maximum pooling method is used to screen the maximum eigenvalue to form a feature fusion vector. The results show that the training set AUC, validation set AUC and test set AUC of the NAZHA-TextCNN model on the clinical medical dataset are 96.35%, 96.12% and 94.63% respectively; the training set AUC, validation set AUC and test set AUC on the clinical medical dataset are 98.35%, 97.12% and 95.63% respectively. It shows that the NAZHA-TextCNN model has high accuracy and strong generalization ability when dealing with multi-label long text classification problems.