The Research on Chinese Text Difficulty Grading Based on Fine-Tuned LLM: A Case Study of Intermediate and Advanced HSK Reading Texts

Text difficulty grading is a fundamental task in natural language processing, aimed at categorizing various texts into different levels of difficulty. Currently, in the field of international Chinese education, the most widely used and effective automatic method for grading is based on the BERT pre-trained model. However, the prediction accuracy of the BERT model has not exceeded 90%. This study employs the Qwen-2 model as the foundational model and utilizes LoRA fine-tuning technology to grade intermediate and advanced reading texts in HSK. The evaluation metrics include accuracy, recall, precision, and F1 score. The experimental results demonstrate excellent performance across all metrics, with accuracy reaching 90%.