基于LERT预训练模型的医疗搜索意图识别研究
Research on Medical Search Intention Recognition Based on LERT Pre-trained Model
曾嘉慧 1田晓琼 1刘超毅1
作者信息
- 1. 湖南省肿瘤医院,湖南 长沙 410013
- 折叠
摘要
正确识别用户意图有助于提高医疗搜索的准确性,为医疗搜索系统的使用者提供便利.为了提高医疗搜索的意图识别准确率,文章利用中文医疗信息处理评测基准中的医疗搜索检索词意图分类数据集,对LERT预训练模型(Chinese-LERT-base)和BERT预训练模型(BERT-base-Chinese)进行了微调,并对微调后模型的意图分类准确率进行了评估.微调后的LERT模型在"治疗方案""疾病表述"和"病因分析"类别的分类准确率较微调后的BERT模型分别提高了4.53%、8%和8.34%,在"其他"类别的分类准确率降低了9.45%,总体分类准确率提高了0.22%.
Abstract
Correct identification of user intention can help improve the accuracy of medical search and provide convenience for users of medical search systems.In order to improve the accuracy of intention recognition in medical search,this paper uses the KUAKE-Query Intent Criterion dataset in the Chinese Biomedical Language Understanding Evaluation to fine-tune the LERT pre-trained model(Chinese-LERT-base)and the BERT pre-trained model(BERT-base-Chinese),and evaluates the intention classification accuracy of the fine-tuned model.The classification accuracy of the fine-tuned LERT model in the"treatment plan""disease description"and"etiological analysis"categories is improved by 4.53%,8%,and 8.34%,respectively,compared with the BERT model after fine-tuning,and the classification accuracy in the"other"category is reduced by 9.45%.The overall classification accuracy is improved by 0.22%.
关键词
信息系统/意图识别/自然语言处理/大语言模型/医疗搜索Key words
information system/intention recognition/Natural Language Processing/Large Language Model/medical search引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据