Chinese Text Classification Using Support Vector Machine Optimized by Adaptive Particle Swarm Optimization

This study introduces an Adaptive Particle Swarm Optimization (APSO) algorithm to optimize Support Vector Machine (SVM) parameters for Chinese text classification. While Chinese text classification has made significant progress, the selection of an appropriate classifier in high-dimensional feature spaces is critical. SVM is widely recognized for its robust performance, but its effectiveness depends on model parameters. APSO, known for its simplicity and efficiency, is employed to optimize these parameters. The research methodology includes text preprocessing, tokenization, vectorization, and feature extraction, followed by SVM training with parameters optimized by APSO. Experiments used a Chinese text dataset, applying HanLP for segmentation, TF-IDF for vectorization, and Chi-square test for feature selection. Different PSO variants were compared in optimizing SVM's γ and C parameters. Results indicate that APSO achieves superior performance, with an F1-score of 84.02%, outperforming other variants. APSO's dynamic adjustment of inertia weight and cognitive coefficients enhances exploration capabilities and stability. This study concludes that the APSO-SVM algorithm effectively improves Chinese text classification. However, limitations exist in APSO's design, suggesting future improvements could focus on more adaptive parameter determination and adjustment mechanisms.