基于自然语言处理(NLP)技术建立化学品危险评估知识图谱的研究  

The research of establishment of knowledge graph of chemical hazard assessment based on natural language processing(NLP) technology

作　　者：刘宝[1] 车礼东[1] 黄红花[1] 郭兵[1] 宋振乾[1] 李红霞 范晓明 董瑞 Liu Bao;Che Lidong;Htmng Honghua;Guo Bing;Song Zhenqian;Li Hongxia;Fan Xiaoming;Dong Rui(Shandong Entry-Exit Inspection and Quarantine Bureau,Qingdao Shandong,266002,China;Weifang Entry-Exit Inspection and Quarantine Bureau,Qingdao Shandong,266100,China;Shandong Product Quality Inspection Research Institute,Jinan Shandong,250100,China;Shandong Food and Drug Administration,Jinan Shandong,250100,China)

机构地区：[1]山东出入境检验检疫局,山东青岛266002 [2]潍坊出入境检验检疫局,山东潍坊266100 [3]山东省产品质量研究院,山东济南250100 [4]山东省食品药品检验研究院,山东济南250100

出　　处：《计算机与应用化学》2018年第7期605-610,共6页Computers and Applied Chemistry

基　　金：国家质检总局科研项目(2016IK207,2017IK227);山东出入境检验检疫局科研项目(SK201734)

摘　　要：化学品危险评估与人民生命财产安全密切相关,化学品信息的运用直接影响危险评估的快捷程度和准确度。建立化学品危险评估知识图谱能有限管理和应用化学品信息。本文对化学品危险评估领域知识进行了梳理,在确定了领域范围、知识内容和基本本体层次关系基础上,提出了一种准确高效的领域知识构建方法—＂NLP及人工智能辅助法＂：首先,利用爬虫技术对数据进行采集和清洗通过数据抓取及数据清洗;从获得7.8亿条结构化数据中利用中文分词、语义分析等技术进行了知识抽取,构建化学品危险评估知识主体层次关系;通过关系映射、语义分析等技术手段抽取本体属性;基于自然语言处理和人工智能技术苟安知识本体及知识图谱。本研究在一定程度解决了化学品评估知识图谱专业性强、数据量大、过程复杂;中文知识图谱构建的开放链接相对缺乏,导致目前国内尚没有成熟的化学品评估中文知识图谱的问题。化学品危险分类信息匹配为场景实现了初步应用,结果表明,化学品危险评估知识图谱在危险分类信息数据查询、匹配验证的应用将人工平均时间从4460秒压缩到137秒,准确率从86.2%提升到94.3%,大幅提高了化学品危险评估数据查询和匹配的效率。本文的工作进一步表明利用知识图谱可以更好的实现行业或专业领域知识的管理,具有重要应用价值。Chemical hazard assessment is closely related to people＇s life and property safety, and the application of chemical information directly affects the rapidity and accuracy of risk assessment. The establishment of a chemical hazard assessment knowledge Map can be limited to the management and application of chemical information. In this paper, the knowledge of chemical hazard assessment is combed, based on the definition of domain scope, knowledge content and basic ontology, this paper puts forward an accurate and efficient method of domain knowledge construction--＂NLP and Artificial Intelligence assistant method＂ ： first, using the crawler technology to collect and clean data through data fetching and data cleaning To extract knowledge from 780 million structured data using Chinese word segmentation and semantic analysis, to construct the knowledge subject level relationship of chemical hazard assessment, to extract ontology attributes by means of relational mapping and semantic analysis, and to supine knowledge ontology and knowledge map based on natural language processing and artificial intelligence technology. This research solves the problem that the Knowledge Atlas of chemical assessment is specialized, the data is large, the process is complex, and the open link of the Chinese knowledge Atlas is relatively scarce, which leads to the lack of mature chemical evaluation of Chinese knowledge map in China. The preliminary application of chemical hazard classification information matching is realized, and the results show that the application of chemical hazard assessment knowledge Map in hazard classification information data query and matching verification will compress the average time from 4,460 seconds to 137 seconds. The accuracy rate increased from 86.2 to 94.3, dramatically increasing the efficiency of data query and matching for chemical hazard assessment. The work of this paper further shows that the use of knowledge Atlas can better realize the management of industry or professional domain know

关 键 词：化学品 危险评估 自然语言处理 知识图谱 

分 类 号：TQ086[化学工程]