Abstract:
In service-oriented architecture, a system is built by integrating multiple existing services. Many service providers offer services featuring similar functionality. Qualities of service (QoS) of these offerings have used by system builders to distinguish services when those services are selected for system integration. When a service provider publishes a service, that provider also specifies the QoS that the service promises to deliver. The reputation of a service reflects how users perceive the extent to which a service delivers on that QoS promised. Therefore, the reputation of a system plays an important role in helping users decide which service to use. A reputation system is employed to record the reputation of the services as reported by the users of those services. As the reputation of services naturally influences whether customers will choose the services to use, it can be envisaged that both malicious users and malicious services will try to manipulate the reputations of certain services in order to gain commercial advantage. Therefore, it is imperative to build a robust reputation system capable of fending off the manipulations made by malicious users and malicious services. Multiple service recommendation systems have been proposed. Many of the existing system do not address the issues that can be caused by the malicious users and services. As a result, these systems are vulnerable to malicious manipulations. This thesis proposes a robust reputation system for recording and protecting the reputation of the services. When calculating the reputation of a service, apart from proposing a reliable personalized reputation model that is built upon a collaborative filtering technique, the system also takes into account the reputation scores that the service is most likely to gain from a group of representative users as well as the experience of the user. Experiments were carried out to evaluate the system. The evaluation shows that (a) the system in this thesis has good predictive accuracy when compared with the existing systems, and (b) the proposed system can effectively counter the manipulations by the malicious users and services.