《電子技術應用》
您所在的位置:首頁 > 其他 > 設計應用 > 基于PCNN的工業(yè)制造領域質量文本實體關系抽取方法
基于PCNN的工業(yè)制造領域質量文本實體關系抽取方法
信息技術與網(wǎng)絡安全
張 彤1,宋明艷1,王 俊1,2,白 洋1
(1.北京京航計算通訊研究所,北京100071;   2.哈爾濱工業(yè)大學 經(jīng)濟與管理學院,黑龍江 哈爾濱150006)
摘要: 對汽車、機械等工業(yè)制造行業(yè)的質量報告進行關系抽取,對于該行業(yè)質量知識圖譜、質量問答系統(tǒng)等研究有著極為重要的意義。針對在工業(yè)制造領域的質量知識圖譜構建過程中尚無公開數(shù)據(jù)集可用的情況,收集了質量文本并進行相應的專業(yè)標注,構建了工業(yè)制造領域質量知識圖譜關系抽取專業(yè)數(shù)據(jù)集?;谠摂?shù)據(jù)集利用分段卷積神經(jīng)網(wǎng)絡(Piecewise Convolutional Neural Network,PCNN)實現(xiàn)關系抽取,然后根據(jù)中文特性,提出了改進的PCNN模型(C-PCNN),以提升在中文語料中關系抽取的性能。在本文構建的數(shù)據(jù)集中,改進后模型的準確率、召回率以及F1值優(yōu)于對比的PCNN和RNN模型,驗證了該方法的可行性和有效性。該研究對從事制造行業(yè)的人員有一定的實際意義。
中圖分類號: TP391
文獻標識碼: ADOI: 10.19358/j.issn.2096-5133.2021.03.002
引用格式: 張彤,宋明艷,王俊,等。 基于PCNN的工業(yè)制造領域質量文本實體關系抽取方法[J].信息技術與網(wǎng)絡安全,2021,40(3):8-13.
Entity relation extraction method for quality text of industrial manufacturing based on Piecewise Convolutional Neural Network
Zhang Tong1,Song Mingyan1,Wang Jun1,2,Bai Yang1
(1.Beijing Jinghang Research Institute of Computing and Communication,Beijing 100071,China; 2.School of Management,Harbin Institute of Technology,Harbin 150006,China)
Abstract: Relation extraction of quality reports in industrial manufacturing industries such as automobiles and machinery is of great significance to the research of quality knowledge graph and quality question answering system of the industry. Aiming at the situation that there is no public dataset available for relation extraction of quality reports in the industrial manufacturing field, this paper collects quality reports in the field of industrial manufacturing and makes corresponding professional labels to construct a professional dataset for relation extraction. Based on this dataset, Piecewise Convolutional Neural Network(PCNN) is used for relation extraction. To be more specific, then based on Chinese characteristics, an improved PCNN model(C-PCNN) based on chinese characteristics is proposed to improve the performance of relation extraction in chinese corpus. Experimental results on the constructed dataset show that the accuracy, recall, and F1 values of the C-PCNN are respectively better than PCNN and RNN, indicating the feasibility and effectiveness of the method. This research has practical significance for personnel engaged in the manufacturing industry.
Key words : industrial manufacturing;quality text;relation extraction;Piecewise Convolutional Neural Network

0 引言

汽車、機械等工業(yè)制造行業(yè)的產(chǎn)品是涉及多個技術領域的高精度、高可靠性產(chǎn)品,具有結構復雜,生產(chǎn)周期長、生產(chǎn)狀態(tài)多等特點[1]。隨著信息化時代的發(fā)展,在生產(chǎn)研制過程中產(chǎn)生的各類質量數(shù)據(jù)日趨龐大,但由于現(xiàn)階段缺乏統(tǒng)一的數(shù)據(jù)管理,各類質量信息散落在業(yè)務系統(tǒng)中,以電子或紙質文檔方式存在,這些離散存儲的質量信息包含各類質量問題的原因、問題部件、采取措施等關鍵信息。如何從這些離散存儲的質量信息中抽取出有效信息,為工業(yè)制造提供數(shù)據(jù)支撐,幫助相關人員有效監(jiān)督產(chǎn)品生產(chǎn)、快速解決質量問題,構成工業(yè)制造領域質量管理的迫切需求。本文從質量文本出發(fā),利用關系抽取技術挖掘文本中實體間存在的語義關系,為后續(xù)構建質量知識圖譜、質量問答系統(tǒng)奠定堅實基礎。




本文詳細內容請下載:http://ihrv.cn/resource/share/2000003422




作者信息:

張  彤1,宋明艷1,王  俊1,2,白  洋1

(1.北京京航計算通訊研究所,北京100071;

2.哈爾濱工業(yè)大學 經(jīng)濟與管理學院,黑龍江 哈爾濱150006)


此內容為AET網(wǎng)站原創(chuàng),未經(jīng)授權禁止轉載。