2-型模糊描述逻辑及其应用研究-智能与分布计算实验室

2-型模糊描述逻辑及其应用研究

姓名	孙小林
论文答辩日期	2007.10.01
论文提交日期	2007.06.22
论文级别	博士
中文题名	2-型模糊描述逻辑及其应用研究
英文题名	The Research of Type-2 Fuzzy Description Logics and Its Application
导师1	卢正鼎
导师2	李瑞轩
中文关键词	专家系统;分布式环境;本体;描述逻辑;2-型模糊描述逻辑;模糊推理;本体进化;数据挖掘;聚类
英文关键词	Expert System;Distributed environment;Ontology;Description Logics;Type-2 Fuzzy DL;Type-2 Fuzzy Reasoning;Ontology Evolution;Data Mining
中文文摘	随着网络技术的高速发展以及信息的迅猛增长，极大丰富了人们可从万维网上获得的资源。许多专家系统期望能够通过这些资源建立庞大的知识库，然而由于绝大部分资源虽然是可获得的却不一定是机器可理解的，如何对分布式环境的资源进行整合并使之具有机器能够理解这些资源的语义成为人工智能技术的新的挑战。为了解决以上问题，新兴的本体技术应运而生。在该技术的支持下，各领域知识通过知识描述，能够方便的整合在一起并被计算机所理解。然而由于本体标准描述语言的支撑逻辑??描述逻辑虽然具有强大的描述能力，但是却无法对现实中的模糊信息进行表达和推理，使得本体知识库无法为模糊专家系统提供支持；同时，本体处在一个开放式的环境中，如何使本体不断地调整自身、不断地进化来适应不断变化的环境也是本体技术研究的热点之一；除此之外，由于缺乏自动建立本体的方法和系统，构建大规模的本体也是各专家系统本体开发人员面临的巨大挑战。已有研究对描述逻辑进行的是1-型模糊扩展，这种方式使用某确定数值来描述隶属度，不仅与现实经验不符也存在诸多限制。在对描述逻辑作了深入的研究后，提出了一种描述逻辑的2-型模糊扩展，详细讨论了2-型模糊描述逻辑的语法、语义等内容，以期使描述逻辑更好地适用于模糊专家系统，并将该模糊描述逻辑应用于语义搜索系统进行实验分析。标准本体建模语言（OWL）能够描述现实世界中的概念及其之间的关系，却无法表达概念之间的因果推理关系，语义网规则语言（SWRL）的诞生提供了对这种本体概念之间的推理规则的定义能力。一个完整的本体知识库的推理机制必须能够支持这种基于规则的推理。将本体术语集和断言集的推理算法进行了模糊扩展，同时以2-型模糊推理技术为基础，提出了一套2-FSWRL本体的推理方法与流程。针对目前绝大多数数据都是保存在关系数据库中的现状，旨在建立一个具有建立大型2-型模糊本体功能的2-型2-型模糊本体进化系统，以数据挖掘技术为基础，提出了一种本体进化框架。为了适应本体数据源大量的分类数据，提出了一种新的聚类算法SCT，实现数据库元组的自动聚类以及概念层次结构的生成。并设计了一套数据库抽取规则，通过对关系模式的分析将关系数据表转换成本体的概念和属性节点。最后以提出的本体进化框架为基础实现了一个2-型2-型模糊本体进化系统：Grampus，主要功能涵盖面向用户的结构驱动本体变化与使用驱动本体变化捕捉与执行、基于数据挖掘技术的数据驱动本体变化捕捉与执行、基于数据挖掘技术与规则的本体源数据导入、面向关系数据库的原始本体构建等方面，并通过系统的试验，给出算法实现的性能分析与评价。
英文文摘	The available resource in WWW increased greatly due to the high-speed development of network and the exploding of infomarion. There are many expert system expect to build an enormous knowledge base with these resources. Most of these resources are available, however, they are hard to be understood by computers. How to integrate these resources and make computers understand the semantic element form them is becoming a new challenge for artificial intelligence. Ontology is proposed to handle the problem mentioned above. Supported by ontology, knowledge from different domain can be integrate together conveniently through knowledge representation and understood by different computers. However, since the describing ability of Description Logics (DLs), the standard supporting logics for ontology, is strong enough, DLs cannot represent or infer the imprecise information around us. Which makes the ontology based on DLs can hardly support the fuzzy expert system. At the same time, Business dynamics and changes in the operating environment often give rise to continuous changes in application requirements that may be fulfilled only by changing the underlying ontologies. There are many researchs toward the type-1 fuzzy extended version of DLs, which describe the imprecise information by a crisp value. Type-1 fuzzy DL cannot accommodate itself to the real-world well and has many limits for knowledge representation. A type-2 fuzzy DL is proposed for this problem. We discuss the syntax and semantic of type-2 fuzzy DL, which is expected to be applied into fuzzy expert system. At last, this fuzzy DL has been implemented in the semantic search engine system to analyse its functions. Ontology Web Language (OWL), the standard language for Ontology modeling, can express the concepts in the real world and the relationship between them. However, OWL can describe the relationship for inference between concepts. Semantic Web Rule Language (SWRL) is introduced to describe the reasoning rules of ontology concepts. That means a whole ontology knowledge base needs to be able to support the reasoning based on rules. With the extend of Tableau algorithm, a complete set of inference flow for type-2 fuzzy SWRL ontology is proposed based on the type-2 fuzzy reasoning technique. For the reason of most information is stored in relational database, an Architecture of ontology evolution is designed based on data mining to build a fuzzy ontology evolution system which can handle fuzzy ontology with large volume data. A new clustering algorithm named SCT is raised, which makes system be more suitable to the category data in data source of ontology. SCT can classify the records in table of relational database automatically, and generate a hiberarchy of concepts. In order to extract information from relational database to ontology, a set of switch rules is proposed. At last , a fuzzy ontology evolution system named “Grampus” is implemented based on architecture introduced above. Its main functions including user-oriented structure-driven change discovery and implementation, data-driven change discovery and implementation based on data mining, ontology data source extraction based on data mining and rules, original ontology building orienting relational database. The above theoretical principles and practical techniques are adopt for developing a prototype. The experiments are carried out to report the evaluation of arithmetics and results of performance analysis.