A multi-attribute service composition model in a dynamic environment

Abstract

Dynamic service composition involves the run-time selection of service elements that are combined to form a larger more complex composite application. There are several issues in dynamic service composition that are examined in this thesis. First, the existence of a large number of service elements with similar or identical functionalities makes it difficult to select appropriate service elements dynamically. To address this issue, we utilize the non-functional attributes of the service elements. The non-functional attribute values are discriminating factors on which one service element is deemed better than other candidates for the specific composition. We first present novel selection techniques based on single attribute criteria for the three non-functional attributes: reliability, waiting-time, and reputation. More generally, however, there are usually a number of service attributes, which include non-functional and functional attributes, associated with each service element. In addition, the customers who are the ultimate stakeholders of the composition process have varied preferences for these service attributes. To incorporate simultaneously all the service attributes and the preference weights of service customers, we introduce a unifying factor called affinity that is a function of all the service attributes of the service elements and the preferences articulated by customers. The affinity factor is embedded in a new model called the Affinity Model that is used for the dynamic selection of service elements to form service compositions. The Affinity Model utilizes the affinity values calculated to select service elements following a Greedy algorithm where the service elements with the largest affinity values are selected for each functionality. The efficacy of the Affinity Model is validated by simulating the service composition process as a game called the Ambitious-Traveler. The validation procedure involves a set of human participants who volunteer to play the Ambitious-Traveler game. The game is simultaneously played in an automated manner using selection decisions made by the Affinity Model. The results show a comparable or superior performance by the Affinity Model to that of the human players in 90% of the trials. This validates the hypothesis that the Affinity Model is capable of making service selections that are comparable to or better than the intuitive judgement of humans.