Degree Name

Doctor of Philosophy


School of Information Systems and Technology


Web services are implementation of Service-Oriented Architecture (SOA) which has been recognized as the next generation framework for the alliance of distributed applications over the Internet. During the last a few years, Service-Oriented Computing (SOC) has emerged to be a significantly important research area that has attracted increasing attention from both the research and industry communities. With the ever-increasing number of services published on the Internet, how to discover and invoke a desired service effectively and efficiently has become a challenging issue. More specifically, the service consumers have become more sophisticated and concern about Quality of Service (QoS) in addition to the functionalities of the services. Therefore, selecting web service with high quality that best satisfies consumer's requirements within a large pool of functionally equivalent service providers is the key factor for web service selection. After the consumers present their requirements, system support is needed to assist the consumers in selecting the desired services automatically or semi-automatically. Furthermore, as individual services with complementary functionalities could be composed together to form a value-added new service, i.e., a composite service, it is a challenging issue to select individual services for the composite service based on the QoS requirements.

The first aspect of the work presented in this thesis is QoS based single service selection. For the success of QoS based single service selection, some critical issues need to be addressed, such as how to express and represent the consumer's QoS requirements more accurately, how to compare and measure the difference between the consumer's requirements and the service's published information, and how to find the best service which not only meets the functional requirements but also fulfils the QoS requirements of the service consumer. The fundamental idea of this work is to measure the similarity between the service consumer's QoS requirements and the service's published QoS information. The service provider with the highest similarity score will be selected and recommended to the consumer. In this thesis, the service request model is defined to represent the consumer's request in the nonfunctional aspect more accurately. A novel similarity computation approach that is based on the relative distance is proposed for calculating the similarity score. The categorical QoS values can be calculated in this approach, based on the set theory and semantic ontology. In addition, an attribute based service customization and selection model is proposed, which extends the attribute based access control policy language to express the additional effects beyond the accessibility. In this approach, different QoS values can be defined and applied to different user groups distinguished by their attributes.

The second aspect of the work presented in this thesis is QoS based composite service selection. In order to satisfy the consumer's requirements, how to select and compose multiple services together as a concrete service composition plan is still a challenge because finding an optimal solution is NP-hard. The key idea is to adopt the Genetic Algorithm (GA) to find a near optimal concrete service composition plan. In this thesis, it is pointed out that the subjective opinion, which could be reflected by the trust, needs to be considered in the selection process. The formal service composition architecture for QoS based composite service selection is presented. The definition of the abstract service composition graph and the concrete service composition graph are provided. A novel Trust-oriented Genetic Algorithm (TOGA) is proposed with the trust evaluation method based on the subjective probability theory. The experiment result shows that the TOGA is able to find a near optimal plan effectively and efficiently. Furthermore, TOGA is extended to identify and support various business relationships among individual service providers. Four basic effect types of business rules among concrete service providers are defined, known as dependency, conflict, positive inference, and negative inference. A formal business rule description language is proposed to support the implementation of service composition. The Business Relationship Matrix (BRM) is captured to represent and handle various business relationships in the service composition.

The major contribution of this thesis is a comprehensive approach to QoS-based single and composite service selection. With this approach developed, it can largely facilitate consumers in selecting right services for their business needs.