基于大语言模型与对比学习的文本可读性自动分级方法研究OA

Text readability greatly affects users'information understanding,absorption,and use.To address the problems of confusion between adjacent readability levels and insufficient high-quality labeled data in Chinese text readability grading,this paper proposes an automatic readability grading method based on large language models and contrastive learning.First,a large language model is used to generate contrastive samples with clear readability gradients to expand the training data.Then,BERT is employed to extract deep semantic features,and contrastive learning is introduced to optimize level boundaries in the feature space,thereby improving the model's ability to distinguish adjacent levels.The experimental results demonstrate that the accuracy(0.880)and F1-score(0.881)of the proposed model outperform all baseline models,suggesting that this method can effectively enhance the performance of automatic readability level classification for Chinese texts,particularly showing a clear advantage in discriminating adjacent-level texts.