【作者】 杨洁；

【导师】 周献中；

【作者基本信息】 南京大学 ， 管理科学与工程， 2014， 博士

【摘要】 人件服务是在智能信息系统领域中目前主流的以人为中心的研究背景下所推动的研究话题。它主要的理念是基于面向服务的架构(SOA架构),将人的思维、智慧等嵌入流程,通过加强服务流程与人的互操作以扩展系统流程对非结构化问题的求解能力。现有的对服务的研究主要是针对软件服务,由此建立的流程对复杂问题的求解能力有限,因而将人嵌入到服务流程以提高其求解能力。同时为了增强对人件服务的理解,人件服务与软件服务间的交互等,语义是人件服务实现所必需的核心技术。本文主要以包含人件服务的服务流程为研究对象,以软件服务的已有技术和人件服务的语义描述为研究基础,以软件服务与人件服务的统一管理、调用(包括检索与选择)为研究内容,通过采用相关的理论与技术,主要包括：Web服务、语义、本体、决策树分类、朴素贝叶斯、TF-IDF算法、命题库、服务质量(QoS)、粒子群优化算法、遗传算法等,以达到提高服务组合流程问题求解能力的目的。本文的主要研究成果与创新点有：(1)提出了一种基于本体的人件服务的形式化描述方法。给出了人件服务的定义,分析了它的特性以及软件服务的已有描述方式,并以本体为描述工具给出了人件服务的形式化方法。给出了人件服务的本体描述中包含的类(class)、属性(property),类之间、属性之间的层次关系,属性的定义域、值域等,并对本体的一致性进行了验证,为服务管理与调用的研究提供了基础。(2)遵循软件服务与人件服务统一管理的思想,采用监督式的分类方法将具有相同或相似功能的服务划分到同一类别中并为它们自动地添加语义标签。分析了服务分类管理的过程,包括服务描述的解析、分类元素的选取、数据预处理、服务的向量化和分类算法的应用。为了提高分类的正确率,应用多种元素组合方式及机器学习算法,并通过实验得出了单分类元素与C4.5决策树分类算法的分类效果更理想的结论。同时,针对人件服务的输出具有长期性、易造成人的负担的特点,使用C4.5算法实现了对人件服务的输出预测。(3)针对服务库中服务数量的快速增长,为了检索到满足用户功能需求的服务并提高服务检索的效率与准确率,以信息检索技术为基础,将语义与语法引入到对服务的检索对象——<documentation>的分析中。在同一服务类别中,提出利用相似度建立服务间关系模型,并对个体服务的描述元素的内容进行语法结构分析,通过语法结构以及语义相似度以完成查询内容与检索服务之间的匹配。(4)为了满足用户对服务与服务流程的非功能需求,将服务选择问题转化为求解目标约束问题。为了符合人件服务的研究特点,提出将信任作为服务的QOS指标之一,并区分了软件服务与人件服务信任值的不同计算方式以及目标函数形式化中软件服务、人件服务对信任不同的依赖性,建立顺序、并行、选择、循环逻辑结构的组合服务QOS模型,采用穷举搜索、随机、遗传算法、粒子群算法进行求解分析,并得出遗传算法在求解效率与正确率上具有优势,人件服务相较于软件服务在选择时更依赖于信任的结论。(5)给出人件服务、软件服务的调用及组合实现流程,并基于项目组开发的服务架构平台,演示了服务的统一分类管理及包含人件服务的组合流程实例。更多还原

【Abstract】 Humanware Service is a topic based on the human-centered web environment in intelligent information system. Based on the service-oriented architecture, humanware service develops system’s solving ability by containing people’s wisdom and enhancing interoperability of people with service process. The process is consisted of software services. But in our process, a node is implemented by humanware service while the task of the node cannot be solved by software. To implement humanware service, semantic is a crucial technique. Because semantic is to enable humanware service understandable and interoperable with software service. In this paper, service management and invocation are studied. Software service and humanware service are stored, managed unifiedly. Service invocation includes two stages:service retrieval and service selection. Based on the research contents, the adoptive theories, methods and techniques include:web service, semantic, ontology, decision tree classifier, Naive Bayes, TF-IDF weighting, propbank, Quality of Service, particle swarm optimization algorithm, genetic algorithm, etc. The main contributions of this paper are concluded as follows.(1) A formal description method of humanware service based on ontology is provided. The definition and feature of humanware service are analyzed. Referring web service’s description OWL-S, ontology as a tool is employed to describe humanware service. The metadata, class, property, et al are listed. Hierarchical relationships of classes, properties and properties’domain, range are defined. Consistency of humanware service descriptional ontology is verified.(2) Service classification, including software service and humanware service, is to classify services with the identical or similar functionalities into one category by applying supervised machine learning technique. And then services will get their own semantic labels automatically. The course of service classification is resolved into: analyzing service description document, extracting element, preprocessing, transforming into a service vector and applying classifiers. For developing classification accuracy rate, single element and C4.5 classifier perform better, which are proven by experiments. The output of humanware service is long-term and overloads people’s mental burden, so the output prediction is achieved by C4.5 classifier.(3) With the number of services in the repository increasing, to search the desired service of user is a meaningful work. To meet user’s functional requirements, based on information retrieval, semantic and syntactic are proposed into service retrieval to develop efficiency and precision. In a service category, it is provided that similarity is used to establish service-related model, and syntactic structure of content of <documentation>——element of service description is analyzed. The return services are matched with query from words’ similarity and sentences’ syntactic structure.(4) To seek quality of service, service selection is transformed into a multi-objective constraint optimization problem. According to humanware service’s characteristic, trust as a quota is provided to measure service as well as reliability, availability, execution time and price. Trust evaluation, uility function in humanware service and software service are provided respectively. Qos model of composed service is introduced, which includes logical structures, sequence, parallel, choice and loop. Exhaustive search, stochastic, genetic algorithm and particle swarm optimization are used to solve the problem. And the experimental results show that genetic algorithm outperforms other three algorithms, and in service selection, humanware service depends more on trust than software service.(5) The invocation process based on software service and humanware service is given. The project group developed a service architecture platform. Unified management of humanware service and software service, command decision process which contains humanware service are illustrated on the platform.更多还原