摘要
针对机械工程领域缺乏标注语料问题,提出一种基于双重深度迁移学习的中文命名实体识别方法。该方法同时将迁移学习应用于预训练语言模型迁移和整体模型迁移,并结合机械工程领域知识进行微调,建立了双重深度迁移学习模型DT-BLC。以齿轮专利为例,利用统计学的方法对识别后的实体进行研究分析。实验结果表明,在小规模数据集上进行机械工程领域命名实体识别时,DT-BLC模型的精确率、召回率、F1分别达到88.98%、92.51%和90.71%,均优于其他模型,且对识别后的齿轮专利实体通过词频分析和新词发现获得齿轮的研究新信息。
In view of the lack of annotation corpus in the field of mechanical engineering,a Chinese named entity recognition method based on double deep transfer learning is proposed.This method applied transfer learning to both pre-trained language model transfer and overall model transfer,and combined the fine-tuning for the field of mechanical engineering,to establish the double deep transfer learning model(DT-BLC).Taking the gear patent as an example,the statistical methods were used to study and analyze the entity after the recognition.The named entity recognition experiments in the field of mechanical engineering were conducted on the small-scale dataset.The results show that the recognition results of DT-BLC model achieve 88.98%in precision,92.51%in recall and 90.71%in F1,which are better than other models.Moreover,the research status and development trend of gears can be obtained from the identified gear patent entities through word frequency analysis and new word discovery.
作者
臧凌玉
张应中
罗晓芳
Zang Lingyu;Zhang Yingzhong;Luo Xiaofang(College of Mechanical Engineering,Dalian University of Technology,Dalian 116024,Liaoning,China)
出处
《计算机应用与软件》
北大核心
2022年第9期219-224,共6页
Computer Applications and Software
基金
国家自然科学基金项目(51775081)。
关键词
机械工程领域
双重迁移学习
深度学习
命名实体识别
专利分析
Mechanical engineering
Double transfer learning
Deep learning
Named entity learning
Patents analysis
