基于多知识库科技报告术语实体链接研究

发布时间：2018-11-27 18:37

【摘要】：科技报告作为一种重要的文献资源,对其的深入挖掘与分析具有重要的价值和意义。然而,目前针对科技报告的研究仍停留在对其基本概念、属性的界定,以及科技报告体系建设研究,而对科技报告内容深入挖掘与分析研究非常少。科技报告中含有大量的专业术语实体,这些专业术语实体多为科技报告的研究主体,代表着我国科学技术的发展现状与未来趋势。因此,对科技报告内容的挖掘分析,识别其中的专业术语实体对推动科技创新具有重要意义。实体识别技术作为自然语言处理的关键技术,可用于自动识别文本中的人名、地名、机构名等实体,将其扩展应用使得自动识别专业术语实体成为可能。本文以科技报告为研究对象,首先利用新词发现技术发现科技报告中未登录的潜在术语新词,然后构建专业术语知识库作为术语实体识别与链接的语料支撑,最后利用Stanford NER实体识别框架实现科技报告中术语实体的自动识别,并与多个知识库进行链接消歧。主要的研究工作如下:(1)针对目前中文分词存在的问题以及科技报告术语的特点,提出了基于词性组合的新词发现方法,通过制定专业术语的词性组合规则抽取符合规则的词串,并根据词串的支持度以及词长、互信息等内外部特征确定新词,有效发现专业术语新词,在一定程度上提高了中文分词的准确度,为术语实体的识别奠定了基础。(2)构建专业术语知识库。实体识别需要大量的语料作为支持,通过训练语料提取实体特征,实现实体的自动识别。由于目前缺乏公开的科技报告术语语料,本文以中国规范术语网提供的专业术语知识作为数据源,利用网络爬虫,数据库等信息技术设计并构建术语知识库。(3)详细介绍了目前实体识别的主流方法,并选择成熟的基于条件随机场模型的Stanford NER开源实体识别框架,通过训练术语实体模型,实现科技报告术语实体的自动识别,并结合多知识库与语义相似度计算实现术语实体的链接消歧。(4)选取国家科技报告服务系统发布的科技报告作为实验数据,设计并开发基于多知识库的科技报告术语实体链接原型系统。该系统主要集成了科技报告数据预处理、新词发现、实体识别与实体链接功能,实现了对科技报告术语实体的自动识别与消歧,并验证了本文方法的正确性和有效性。
[Abstract]:As an important document resource, it is of great value and significance to excavate and analyze the scientific and technological report. However, at present, the research on science and technology report is still focused on its basic concept, definition of attributes and construction of science and technology report system. There are a large number of technical terminology entities in science and technology reports, which are the main research subjects of science and technology reports, which represent the development status and future trend of science and technology in China. Therefore, it is of great significance to excavate and analyze the contents of science and technology reports and identify the technical terminology entities. As the key technology of natural language processing, entity recognition technology can be used to automatically recognize the names of persons, place names, agency names and other entities in the text. In this paper, the scientific and technological report is taken as the research object. Firstly, the new term discovery technology is used to discover the potential new term in the scientific and technological report, and then the specialized terminology knowledge base is constructed as the corpus support for the identification and link of the term entity. Finally, the Stanford NER entity recognition framework is used to realize the automatic recognition of the terminology entities in the scientific and technological reports, and links disambiguation with multiple knowledge bases. The main research works are as follows: (1) aiming at the problems existing in Chinese word segmentation and the characteristics of the terms in scientific and technological reports, a new word discovery method based on part of speech combination is proposed. By drawing up the rules of part of speech combination of professional terms to extract the words in accordance with the rules, and according to the support degree of the strings and the internal and external characteristics of the words, such as length and mutual information, the new words are determined, and the new words of the professional terms are found effectively. To some extent, it improves the accuracy of Chinese word segmentation, and lays a foundation for the identification of terminology entities. (2) constructing the specialized terminology knowledge base. Entity recognition needs a large number of corpus as the support, through training corpus to extract entity features to achieve automatic entity recognition. Due to the lack of public scientific and technological reporting terminology data, this paper uses the technical terminology knowledge provided by the China Standard terminology Network as the data source and uses the web crawler as the data source. Database and other information technologies design and construct the term knowledge base. (3) the main methods of entity recognition are introduced in detail, and the mature Stanford NER open source entity recognition framework based on conditional random field model is selected to train the term entity model. Realizing the automatic recognition of the technical report term entity, and combining the multi-knowledge base and semantic similarity calculation to realize the link disambiguation of the term entity. (4) selecting the science and technology report issued by the national science and technology report service system as the experimental data. This paper designs and develops a prototype system of entity link of scientific and technological reporting terms based on multi-knowledge base. The system mainly integrates preprocessing of scientific and technological report data, neologism discovery, entity identification and entity link function, realizes automatic recognition and disambiguation of scientific and technological report term entity, and verifies the correctness and validity of this method.
【学位授予单位】：华中师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：G353.1

【相似文献】