随着移动互联网技术的日趋成熟与完善，移动应用已成为用户接入互联网、获取网络服务的主要手段。服务组合技术在移动领域广泛应用，不但使应用开发更为快捷，而且使用户可以在一个应用中访问到不同服务，从而带来更佳用户体验。然而，现有的服务组合采用的是一种静态的、分类聚集的组合方式。应用中组合哪些服务，以怎样的方式组合，在应用开发时就已确定，且在应用运行过程中保持不变。为了满足更多用户需求，应用开发商往往倾向于在应用中集成更多服务，很容易导致应用功能过载、服务索引层次过深等问题。而分类聚集的组合方式使服务间的关联过于薄弱，选择哪些服务执行，以怎样的顺序执行完全依赖于用户的主观经验，过多的服务功能以及过深的服务索引严重影响了用户选择和使用服务的效率。

针对上述问题，本论文提出以“用户为中心”的动态、按需服务组合，根据每个用户的目标及所处的上下文环境，动态构造服务组合逻辑，明确用户对服务的功能需求，然后基于用户偏好与服务质量（QoS）实现高效的动态服务绑定。具体研究内容如下：

（1）面向移动终端用户的目标分析与定义。针对终端用户目标描述不具体、不完整的问题，提出基于历史的目标分析与定义方法，通过解析大众用户访问服务的历史记录构建目标本体，实现基于目标本体的用户目标分析。该方法更适合移动应用中非流程式的应用场景（即服务间不存在严格的流程控制及数据依赖关系）。目标定义以“与或树”为定义模型，支持对“包含”、“与”、“或”等关系的定义，与现有的无层次结构的目标定义模型相比，语义表达能力更强，使基于目标的服务匹配更加准确。

（2）上下文感知的探索式服务组合。针对当前移动应用中服务组合灵活性差、服务访问效率低的问题提出上下文感知的探索式服务组合方法，在用户使用应用的过程中根据所处的上下文环境动态构造服务组合逻辑，使服务组合更能满足用户的个性化需求。该方法主要面向移动应用中非流程式的应用场景，通过挖掘用户在不同上下文环境下访问服务的历史记录，得出上下文与服务间的关联规则，实现基于关联规则的服务推荐，为用户选择服务提供帮助与支持。

（3）基于服务功能规约的动态服务绑定。面向实时、动态的服务选择场景提出基于服务功能规约的高效的动态服务绑定方法，根据用户偏好、服务质量等属性实现多目标多约束条件下的服务组合优化。该方法在传统抽象服务建模的基础上，利用移动应用中服务间的集合式组合特征，将功能相关的抽象服务规约为一个粒度更大的抽象服务，然后在规约后的抽象服务组合上实施组合优化，从而减小服务组合规模及候选服务规模，提高组合优化效率。

As the mobile Internet technology becoming mature and perfect, mobile application has become the primary way most people experience the Internet and access to web services. Service composition is fully practiced in mobile app. It not only quickens the app development, but also enables users to access to different services in just one app, and then brings a better experience. However, in current apps, most services are composed in a static and clustering-based way. What to compose and how to compose are all determined in app development stage and will not change when the app is running. To fulfill more user requirements, app providers are always prone to integrate more services in one app. This can easily lead to functionality overload and overlong service index path. Furthermore, clustering-based composition weakens the relationship between services and the service execution path is mainly determined by the user himself. Too many functionalities and too long index path have a great damage to the efficiency of service accessing.

In this thesis, we propose a user-centric method for service composition which composes services in an on-demand and dynamic manner to overcome above problems. Based on the current user goals and context, one or more abstract services describing the functional user requirements are composed into the app just in time. And then, a set of concrete services matching the desired functional requirements are picked out and bound to those abstract services according to user preference and the quality of services. Research issues discussed in this thesis are presented as followings:

(1) Analysis and definition of mobile user goal. In the case of non-specific and incomplete goals given by user, we present a history-based method to help users to analyze and define their goals. We build a goal ontology based on the history records which record the traces of service call made by different users. According to the goal ontology, non-specific goals are decomposed into specific goals and incomplete goals are supplemented to be complete. This method is suit for non-process-like app scenario which means there is no definite control flow and data flow between services. We define user goals based on an and/or tree model which enables the semantic expression such as contain, and, or. Compared with the current flat definition model, our model is more expressive and makes service matching more precise.

(2) Exploratory service composition based on context. In the case of static service composition and low efficiency of service access, we present an exploratory and context-aware method for service composition which composes services step by step based on the context when the user using the app. This makes the composition more personalized and satisfied. This method is mainly used for non-process-like app scenario. Based on the mining of history records which record the service access in different context by different users, a set of association rules between the context and service can be derived, and then rule-based service recommendation is used to help users to determine what kind of services they should use.

(3) Dynamic service binding based on functionality folding. To cope with real time and dynamic service selecting and binding, we present an efficient and functionality-folding based method to realize multi-objective and multi-constrained combinatorial optimization for services according to user preference and the quality of service. Based on the traditional abstract service model, an operation named functionality-folding is defined to combine some of the original abstract services into a coarse-grained abstract service according to the functionalities they describe. Combinatorial optimization is then carried out on the folded abstract services. After folding, the scale of both abstract services and candidate services are greatly reduced, and then the whole algorithm is accelerated.