An Efficient Training Framework for Chinese Text Classification: Integrating Large Language Models and Reinforcement Learning

This study proposes an efficient model training framework aimed at improving the performance of hierarchical classification of Chinese text in low-computation environments. By leveraging the generative capabilities of large language models for data annotation and augmentation, a high-quality dataset is constructed. Knowledge distillation techniques are applied to transfer knowledge from a complex teacher model to a lightweight student model, effectively reducing computational resources. Additionally, a reinforcement learning strategy is introduced to further optimize the student model, enabling it to surpass the teacher model in both accuracy and inference efficiency, providing a novel solution for natural language processing and other resource-constrained tasks.