PhD Defence - Abdullah Abdullah

Title: Towards a Comprehensive Web Service Recommendation Framework

Abstract:

Web services nowadays are considered a consolidated reality of the modern Web with remarkable, increasing influence on everyday computing tasks. They provide standardized means of publishing diverse, distributed applications, turning the Web into major, most acknowledged distributed computing platform ever. Following Service-Oriented Architecture (SOA) paradigm, corporations are increasingly offering their services or programs within and between organizations either on corporate intranets or on the cloud. The list of SOA application domains includes e-business, e-government, automotive systems, multimedia services, finance, process control, etc.

While each Web service provides its own functionality, many of them can be composed together to achieve more complex functionality, i.e., value-added service. Further, with increasing adoption of Web services on the Web, Quality-of-Service (QoS) is becoming important metrics for describing nonfunctional characteristics of Web services. Therefore, the aim of this work is to advance the academic efforts in assisting end users and corporations to benefit from Web service technology by facilitating the recommendation and integration of Web services into composite services.

In this thesis, we propose a recommendation framework that is capable of not only recommending an individual Web service but also a composite one when no service available to fulfill the user request. The framework is realized into two main parts: first a recommendation model for individual Web service is proposed where the QoS profile is considered as an implicit rating scheme. The model utilizes the Jaccard coefficient in several variants to create two Unipartite similarity-based graphs that capture similarities among Users and among Services. By integrating these graphs with the original user-service rating graph, a richer recommendation model is constructed. Using the Top-K Random Walk recommendation algorithm, a final set of recommendations is delivered to end user. The model proves its well-behaviour in terms of sparsity tolerance and recommendation accuracy.  To minimize complexity of the proposed model, a thresholding technique is proposed in which Random Walk algorithm is better guided using a reduced subset of users based on their Jaccard similarities, rather than the entire set. Furthermore, the applicability of the proposed model as a generic recommendation model in also examined through the use of an ordinary rating domain.

The second component is a service composition model in which AI-based planning using Agent technology is adopted to dynamically and flexibly construct composite service workflow. In this model, a distributed service dependency model based on AND/OR graph structure is decomposed and distributed among individual members of the Agent community. The agents are equipped with a well-defined internal reasoning mechanism based on agents' knowledge. Using a communication protocol, the agents actively collaborate to find a cost-effective executable workflow according to end user request. Finally, feasibility and effectiveness demonstration of all components of the proposed framework, using publicly available datasets, a recommendation library, and a multi-agent platform is verified.

Chair: Dr. Mark Wineberg
Advisor:  Dr. Xining Li
Advisory Committee Member: Dr. Stefano Gregori
Non-Advisory Committee Member: Dr. Fangju Wang
External Examiner: Dr. Jianguo Lu [University of Windsor]