基于BERT-BiLSTM-TFIDF的产品研发文档关键词抽取方法  

Keyword Extraction for Product Research and Development Documents Using BERT-BiLSTM-TFIDF

作　　者：卢啸岩 郑宇[1] 昝欣 LU Xiaoyan;ZHENG Yu;ZAN Xin(School of Mechanical and Power Engineering,Shanghai Jiaotong University,Shanghai 200240,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240

出　　处：《工业工程与管理》2023年第2期99-106,共8页Industrial Engineering and Management

基　　金：国家科技支撑计划课题(2015BAF18B00);国家自然科学基金资助项目(51505286);国家工信部智能制造专项(MC-201720-Z02)。

摘　　要：制造企业现有的内部知识管理系统大多通过人工选取产品研发文档标签,效率低下。应用自然语言处理技术抽取文档关键词作为文档标签有助于制造企业知识管理系统智能化。针对产品研发文档关键词抽取问题,提出了BERT-BiLSTM-TFIDF关键词自动抽取方法,基于BERT-BiLSTM设计句权重模型计算各词语所在句子的句权重,同时添加词性权重以及外部语料库以改进TFIDF算法。本文提出的方法改善了现有关键词自动抽取方法没有合理利用词语的语义信息、上下文关系信息的缺点,经过实验证实具有较好的效果。Most of the existing internal knowledge management systems of manufacturing enterprises select the document labels manually,which is inefficient.Extracting keywords automatically to generate document labels using natural language processing technology contributes to the intelligentization of the knowledge management system.For the keyword extraction of automobile research and development documents,this paper proposed the BERT-BiLSTM-TFIDF keyword extraction model.This proposed model added sentence weights and external corpus to improve TFIDF method.The sentence weights were calculated with a designed BERT-BiLSTM model.The proposed keyword extraction method has improved the shortcomings that the existing keyword extraction methods could not make use of the semantic information and context of the word.The proposed BERT-BiLSTM-TFIDF method achieves a good result through experimental verification.

关 键 词：关键词抽取 产品研发文档 BERT-BiLSTM-TFIDF 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术] TH122[自动化与计算机技术—计算机科学与技术]