层次分析决策方法与网页信息抽取技术研究
文献类型:学位论文
作者 | 吴龙庭 |
学位类别 | 工学博士 |
答辩日期 | 2009-05-30 |
授予单位 | 中国科学院研究生院 |
授予地点 | 中国科学院自动化研究所 |
导师 | 戴汝为 |
关键词 | 决策支持 综合集成研讨厅 层次分析法 网页信息抽取 数据挖掘 Decision support HWME AHP Web Information Extraction Data mining |
其他题名 | Research on the Analytic Hierarchy Process and Web Information Extraction Techniques |
学位专业 | 模式识别与智能系统 |
中文摘要 | 综合集成研讨厅是以钱学森为首的一批中国科学家于二十世纪九十年代初提出的 致力于于解决开放的、复杂的、具有多方面不确定因素的宏观决策问题的智能决策系统。它集合了计算机、信息、决策和思维等科学 领域最新的研究成果,以人为主,人机结合,在经济、军事、教育和医药决策 领域具有重大的应用价值。本文从丰富和发展综合集成研讨厅的理论体系和应用技术入手,研究层次分析法(AHP)和网页信息抽取技术;通过引入标度误差和判断 误差,提出AHP中减少判断数量对计算排序向量影响的问题;为解决AHP排序翻转现 象,提出AHP迭加法则合理性检验的四项标准;分析AHP庸准则消除问题,提出消除 庸准则同时保持方案排序的方法;提出网页信息抽取方法,设计使用JavaCC开发网页信息 抽取器的技术路线;最后基于天涯论坛进行了社区兴趣发现和社区挖掘工作,为将互联网纳 入综合集成研讨厅的信息采集源作了初步准备工作。 本文首先介绍了综合集成研讨厅、层次分析法和网页信息抽取的基本概念,阐述了它们 的研究概况和主要的研究内容,并对本文的选题背景和主要内容作了介绍。 其次,根据决策科学的发展历史,指出AHP产生的具体背景;阐述了AHP使用的一般步骤, 包括建立递阶层次结构、构造成对比较判断矩阵、计算排序向量和计算合成权重,最后 论述AHP方法的主要特点。 第三,研究层次分析法领域的三个理论问题,分别是判断数量消减、迭加法则检验和 庸准则消除。提出了AHP中的判断数量消减问题,在AHP中引入标度误差和判断误差概念, 推演出新的成对比较判断矩阵,指出AHP中使用1-9标度的原因。在引入新[1-9]标度代替 1-9标度的基础上,提出新的排序向量计算方法。证明了减少判断数量会降低决策可靠性 的第一个特例。针对AHP中的排序翻转现象,提出了四个标准用于检验AHP中迭加法则的 有效性。使用该四种标准验证三种常用的AHP迭加法则,证明了仅乘积AHP法符合所有四种 标准。研究了AHP中的庸准则消除问题,提出了在AHP中消除庸准则的方法。 第四,介绍网页信息抽取技术的基本概念和使用方法,提出从网页中抽取信息的 基本技术,使用编译器生成工具JavaCC开发出网页信息抽取模块,实现了对强国论坛的 网页信息抽取。 第五,抽取天涯虚拟论坛的网页信息,对抽取信息进行数据挖掘和知识发现,研究论坛网友 的社会兴趣和社区结构。 最后,对本论文所进行的工作和取得的成果加以总结,并指出需要进一步做的工作。 |
英文摘要 | The Hall for Workshop of Metasynthetic Engineering (HWME), which is proposed by a group of Chinese scientis in early 1990s, aims to resolve marcroscopical decision-making problems with open, complex and dynamic properties. Integrating the latest achievements of computer, information, decision and cognitive sciences, emphasizing human's influence and adopting human-machine systems, HWME is of substantial value to economical, military, educational and medical decision-making sector. Aiming to enrich HWME technologies and develop its theoretical system, the dissertation studied the Analytic Hierarchy Process (AHP) therory and Web information extraction (IE) technology. Through introducing scale error and judgment error, the dissertation proposed judgment reduction issue in the AHP. To resolve rank reversal phenomenon, it proposed four criteria to validate AHP aggregation rules. At last it used JavaCC to generate Web information extractor and did data mining experiments to discover forum social interests and forum commnunity. The job made preparations for the inclusion of Web into HWME. Firstly, the basics of HWME, AHP and IE are intorduced. The research development and main research directions are reviewed. The background and structure of this thesis are also addressed. Secondly, according to the history of the development of decision science, the background in which AHP is created is stated. The procedures of using AHP to solve decision-making problems are elaborated, including designing hierarchies, building pair-wise comparison matrix, computing priority vector and calculating overall priorities. At last we concluded the main features of AHP. Thirdly, three AHP theoretical problems, namely judgment number reduction issue, aggregation rule evaluation and wash criteria problem, are studied. We proposed judgment number reductiion issue in AHP. By bringing the concepts of scale error and judgment error, a new form of pair-wise comparison matrix is deduced and the reason why AHP uses 1-9 scales is explained. Based on using [1,9] scales to replace former 1-9 scales, new method to calculate priority vector is proposed. We prove the first example in which reducing judgment number would harm the reliability of AHP decision. Aiming at resolving the rank reversal disputes in AHP, we introduced four criteria to evaluate the validity of the AHP aggregation rule. Wash criterion issue is studied and a method used to eliminate wash criteria in AHP is stated. Fourt... |
语种 | 中文 |
其他标识符 | 200618014628031 |
源URL | [http://ir.ia.ac.cn/handle/173211/6193] ![]() |
专题 | 毕业生_博士学位论文 |
推荐引用方式 GB/T 7714 | 吴龙庭. 层次分析决策方法与网页信息抽取技术研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2009. |
入库方式: OAI收割
来源:自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。