【作者】 于红；

【导师】 王秀坤；

【作者基本信息】 大连理工大学 ， 计算机应用技术， 2006， 博士

【摘要】 随着Internet技术的发展，Web上各种在线信息源不断涌现，这些信息源种类繁多，结构各异。在互联网这样一个动态的环境中，各信息源不断地发生着变化，在这样一个动态、异构、开放的环境下快速、准确地获取信息是十分困难的。因此，研究信息集成系统，为用户提供一个访问异构数据源的统一接口有着非常重要的意义。在这种背景下，本文对信息集成系统的主要技术进行了深入的研究，具体研究工作如下： 信息集成系统体系结构是研究系统的组成模块以及各模块之间关系的。本文对目前几种典型的信息集成系统体系结构进行了分析，总结了这些体系结构的优点和不足。在对实际的信息集成工作中遇到的问题及其他相关需求进行分析的基础上，提出了综合信息集成系统体系结构。该体系结构既考虑对已有投资的保护，又考虑对不断生成的新数据的有效处理；既考虑对传统数据库中的数据和XML数据的处理，又考虑信息检索和辅助决策的需求。是一个面向目前大多数企业的综合信息集成系统。 模式映射的建立是信息集成系统的重要环节之一。本文提出了基于划分的映射模式发现框架——PBMSDF(Partition Based Mapping Schema Discovery Framework)。Dhamankar等提出了iMAP框架，该框架采用搜索器集合和Beam Search的方法，可以自动发现1：1映射、1：n和n：1的复杂映射，但是存在以下不足：一是不能发现m：n的复杂映射；二是该框架需要对属性及属性值实例进行分析，导致系统开销很大；三是对于Web信息集成，获取属性值的数据实例有时是不可能的，因此该框架不能用于Web信息集成。He等提出了DCM(Dual Correlation Mining)框架，该框架采用对属性在模式中的分布进行分析发现属性之间相关度的方法来发现属性之间的映射关系，可以解决iMAP框架中存在的问题，仍存在以下不足：一是由于该框架采用的相关度衡量标准对部分属性的评价结果不准确，导致发现结果准确性不高；二是该框架中采用的AprioriCorrmining和DualCorrelationmining挖掘算法在由相关e项集生成相关e+1项集的过程中，将整个属性集合中的每一个属性添加到相关e项集中构成候选相关e+1项集，然后再判断该e+1项集是否相关，增加了许多不必要的计算，致使搜索空间过大，算法的效率较低。本文提出了比较适合于衡量属性之间相关度的C-衡量标准，并在PBMSDF框架中采用了该标准以提高发现结果的准确度；提出了基于划分和栈的模式映射发现算法，理论分析和实验结果均证明，该框架较iMAP和DCM具有较高的性能和较好的模式发现能力。 XML成为目前Web上数据表示和信息交换的标准，处理XML数据是集成系统的功能之一，XML查询的效率将直接影响集成系统的性能。本文提出了一种基于递归模更多还原

【Abstract】 With the development of Internet, all sorts of online information sources are increasing rapidly. The type and structure of these information sources are different. The structure and content of the information sources keep changing. It will be very difficult to retrieve information from the dynamic, heterogeneous, open circumstance. It is very significative to research the information integrated system and provide a uniform interface for the users. Under such circumstances, the main techniques of information integration are discussed. The key contributions are as follows.The architecture show the modules of the information integrated system and the relations among the modules. The famous architectures of information integrated system are analyzed. The advantages and disadvantages of the architectures are discussed. On the basis of analyzing the requirement of the actual information integration system and other ralative information integrated system, the architecture of synthetical information integration system is proposed. In the new architecture, both the existent information and the producing information are handled, the data from traditional database and XML Schema are managed, the information retrieving as well as the assistant decision are considered. It can meet the requirement of most enterprises.The creation of the schema mapping is one of the importment steps in information integrated system. The PBMSDF (Eartition Based Mapping Schema Discovery Eramework) is presented to efficiently discover the mapping schema. Dhamankar proposed iMAP frame, in which the set of searcher and Beam Search are applied. 1:1, 1:n and n:1 mappings can be discovered. However, there are some disadvantages in iMAP frame. Firstly, the m:n mapping can’t be discovered. Secondly, the attributes and the instance of the attributes must be analyzed to discover the maaping schema so that the cost of the system is expensive. Finally, geeting the instance of attributes is impossible for Web information integration. Therefore, the iMAP frame can’t be used in Web information integrated. He proposed DCM (Dual Correlation Mining) frame, in which the problems of iMAP frame are resolved. However, the Shortcomings of DCM are as follows. On the one hand, the measure of the correlation is inaccurate sometimes so that the the result is unauthentic on occasion. On the other hand, the time cost of AprioriCorrmining and DualCorrelationmining mining algorithms has a lot of unnecessary searching. In this paper, The C-Measure is proposed, the partition and stack based attribute group and mapping schema discovery algorithm are proposed. The former can calculate the correlation among attributes and the latter can reduce the search time. The更多还原