【作者】 张程；

【导师】 韩燕波；

【作者基本信息】 中国科学院研究生院（计算技术研究所） ， 计算机软件与理论， 2006， 博士

【摘要】 作为一种新兴的、面向Internet的分布式计算模式，面向服务的计算为构造松耦合、跨组织的集成应用提供了更好的使能技术。面向服务架构通过发布-查找-绑定的模式为使用服务资源提供了基本保障，但是服务所具有的使用者与提供者相分离的特征增加了用户理解、得到及使用所需服务的难度。特别是当用户的需求随着应用构造过程的演进而变化时，如何让用户及时的得到合适的服务是一个需要解决的问题。本文从方便用户使用的角度出发，以面向服务架构作为基础，提出一种服务的个性化推荐方法，以“随需而现”的方式主动的向用户推荐符合其当前需求的服务。同时，围绕其中的核心内容从相关模型、关键技术等方面进行了研究，其主要工作和创新性贡献如下： 提出一种基于动态标价的服务推荐方法 与已有的相关工作相比，该方法的特色体现在：针对构造应用的过程不断演进、并且完成推荐时用户还没有提供显示需求的特点，将服务推荐转化为实时计算服务标价并对服务动态排队的问题。其核心在于以当前语境来刻画隐含在应用构造过程中并且不断变化的隐式用户需求，并根据语境信息计算候选服务的标价，进而对服务排序。该方法的效果体现在：构造应用的每个步骤中，实时的计算候选服务的排列顺序，得到符合当前语境的服务集合，主动呈现给用户，从而实现服务资源的“随需而现”。 创建了一种从多角度刻画隐式用户需求的多元语境模型 与已有的只考虑用户因素或只考虑服务因素的语境建模方法不同，该模型从应用构造状态和用户所处环境两个角度刻画了隐含在应用构造过程中、并且不断变化的用户需求形成当前语境。模型中定义了语境信息所包含的基本元素及其计算方法。该模型的一个特点是：语境信息的具体值通过计算生成，当构造应用的过程发生演进时，可据此更新当前语境，从而反映最新的隐式用户需求。 提出了用于完成标价计算的多元语境信息的使用方法 针对语境信息多元化的特点，该方法通过两阶段计算的方式完成服务标价，首先使用不同种类的语境信息分别计算候选服务的标价值，然后对多个标价值做处理形成综合标价。解决的难点问题主要包括： (1) 使用当前语境中所包含的用户兴趣信息计算服务标价时，提出了最优婚配算法用于解决计算过程中所遇到的对元素之间存在多对多关系的两个向量计算一一对应的最大匹配值的问题。与已有算法相比，最优婚配算法避免了冗余计算。同时文中还对算法的正确性给出了证明。 (2) 使用当前语境中所包含的用户习惯信息计算服务标价时，解决了三个难点问题分别是：1．提出了基于有向图的服务调用过程模型及相应算法来生成用户习惯，它是计算服务标价的依据；2．提出了对服务调用过程模型进行转化的更多还原

【Abstract】 As one of the emerging internet-oriented patterns for distributed computing, Service-Oriented Computing (SOC) offers better enabling technologies for integrating loosely-coupled and cross-organizational applications. Though Service-Oriented Architecture (SOA) ensures service resources can be utilized by employing a publish-find-bind pattern, it is still difficult for users to understand and use required services directly. Our concern is how to enable the user to get right services among many possible choices. From business users’ perspective, we propose an approach for personalized services recommendation in this thesis, with which services suitable for a given set of requirements are recommended automatically, and the service set varies at runtime as the requirements change. Major contributions of the thesis include: An approach to runntime service recommendation based on dynamic rankingThe problem of service recommendation is transformed to the problem concerning how to sort service dynamically through real-time ranking, which is more suitable for the case that user’s requirements are subject to changes during applications constrction and execution. In this approach, we depict the changing requirements implied in the process of constructing applications by user’s current situations, which are also the bases for computing services’ ranks. The effectiveness of this approach lies in that in each step of application construction, the order of candidate services is determined by real-time calculation according to both services’ and users’ situations. A muti-modal situation model covering users’ implicit requirementsDifferent from existed situation models that only take either users’ factors or services’ factors into account, this model depicts the user’s requirements that are continually changing and implied in constructing applications from two perspectives, the status of application construction and the environment the user is in. The model defines basic elements included in situation information and the related computing method for each of the elements. One of the most important characteristics of this model is the value of situations is updated timely to reflect the user’s latest requirements when the constructing process of applications evolves. A method for computing services’ ranks based on muti-modal situationsWe compute services’ ranks by two-phase styles because of the muti- meta feature of the current situation. When a service is ranked, various situation information is used respectively to figure out muti-ranks for this service, then these ranks are integrated into a synthetical更多还原