【作者】 王雪；

【导师】 谢晓东；

【作者基本信息】 华中科技大学 ， 计算机软件与理论， 2012， 硕士

【摘要】 随着领域本体的作用受到越来越多的领域的认可，本体技术在知识管理和语义网中扮演着愈加重要的角色，但是领域本体真正被运用到实际应用领域还有几个问题需要克服。其中一个关键问题就是概念、关系识别，尤其在大型复杂的应用领域，概念及关系的识别变得更加耗时耗力，而且由于不同的人对概念的理解不同使得对概念的定义引起一定的争议。为了减少本体构建的耗费，各界提出了领域本体自动、半自动构建，主要是利用机器学习技术、统计方法或是自然语言处理技术从相关的数据源中自动、半自动的进行概念、关系抽取，从而完成领域本体的自动、半自动构建。通过参考相关叙词表和领域文档，提出一种运用相关技术进行概念、概念间关系抽取来完成航空领域本体的半自动构建机制。实现自动化本体构建的关键技术包括概念的抽取、概念间等级关系及非等级关系的抽取方法在文中都有涉及。概念的获取，主要参考相关的叙词表获得核心概念集合。关于概念间等级关系的获取，主要基于现有领域相关叙词表的分类体系作为基础，参考叙词表并利用改进的层次聚类算法提取概念间的等级关系，对原有等级分类体系进行丰富扩充。概念间非等级关系的获取，采用关联规则和模板相结合的方法提取非等级关系，利用支持度、置信度等度量，对经过分词、文档向量提取的文档获取相关概念，利用汉语句法特点和制定好的规则模板，获取概念关系。这样不单减少统计方法缺乏语义逻辑的不足，也降低了概念间语义关系过分依赖语言处理的弊端。最后通过Jena的API对提取的概念关系进行本体结构化等各项操作生成OWL本体，并使用斯坦福大学的Protégé工具进行OWL本体的可视化管理。更多还原

【Abstract】 Though the utility of domain Ontologies is now widely acknowledged in anincreasing number of domains, Ontology is playing a more and more important role inknowledge management and the Semantic Web. Several barriers must be overcome beforeOntologies become practical and useful tools. A critical issue is the task of identifying,defining the concept definition. In case of large and complex application domains thistask can be lengthy, costly and controversial (since different persons may have differentpoints of view about the same concept). To reduce cost, several automated orsemi-automated construction approach have been proposed, which use machine learningtechnologies, statistic approaches or natural language processing techniques for conceptextracting and relation extracting from existing data sources so as to construct domainontology automatically or semi-automatically.This paper presents a Arisa-based domain ontology construction mechanism toextract domain ontology from thesaurus and domain-related text documentssemi-automatically. The key technologies of ontology construction such as domainconcept extraction and relations extraction are discussed in this paper, includingtaxonomic relations and non-taxonomic relations between concepts. The core concepts canbe got or converted form related thesaurus. Taxonomic relations can be extracted usingimproved concept hierarchical clustering approach referred to related thesaurus.As fornon-taxonomic relations extraction,we choose pattern-matching approach and associationrules to extract from unstructured text documents.Use support and confidence to attainrelated items occurred between the documents which have been proceed by ICTCLASincluding tokenizer and lexical analysis. Additionally, gain relations accurately withdetermined templates considering Chinese syntax.Finally, use Jena API to formalize the concepts and relations in order to generate theOWL ontology,then use Protégé to realize visualization management of OWL ontology.更多还原