第28卷第2期2023年4月工业工程与管理IndustrialEngineeringandManagementVol.28No.2Apr.2023基于BERT-BiLSTM-TFIDF的产品研发文档关键词抽取方法卢啸岩,郑宇*,昝欣(上海交通大学机械与动力工程学院,上海200240)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。关键词:关键词抽取;产品研发文档;BERT-BiLSTM-TFIDF中图分类号:TP391.1;TH122文献标识码:AKeywordExtractionforProductResearchandDevelopmentDocumentsUsingBERT-BiLSTM-TFIDFLUXiaoyan,ZHENGYu*,ZANXin(SchoolofMechanicalandPowerEngineering,ShanghaiJiaotongUniversity,Shanghai200240,China)Abstract:Mostoftheexistinginternalknowledgemanagementsystemsofmanufacturingenterprisesselectthedocumentlabelsmanually,whichisinefficient.Extractingkeywordsautomaticallytogeneratedocumentlabelsusingnaturallanguageprocessingtechnologycontributestotheintelligentizationoftheknowledgemanagementsystem.Forthekeywordextractionofautomobileresearchanddevelopmentdocuments,thispaperproposedtheBERT-BiLSTM-TFIDFkeywordextractionmodel.ThisproposedmodeladdedsentenceweightsandexternalcorpustoimproveTFIDFmethod.ThesentenceweightswerecalculatedwithadesignedBERT-BiLSTMmodel.Theproposedkeywordextractionmethodhasimprovedtheshortcomingsthattheexistingkeywordextractionmethodscouldnotmakeuseofthesemanticinformationandcontextoftheword.TheproposedBERT-BiLSTM-TFIDFmethodachievesagoodresultthroughexperimentalverification.Keywords:keywordextraction;productresearchanddevelopmentdocuments;BERT-BiLSTM-TFIDF文章编号:1007-5429(2023)02-0099-08DOI:10.19495/j.cnki.1007-5429.2023.02.011收稿日期:2022-06-22基金项目:国家科技支撑计划课题(2015BAF18B00);国家自然科学基金资...