Agent-based service selection

The current infrastructure for Web services supports service discovery based on a common repository. However, the key challenge is not discovery but selection: ultimately, the service user must select one good provider. Whereas service descriptions are given from the perspective of providers, service selection must take the perspective of users. In this way, service selection involves pragmatics, which builds on but is deeper than semantics. t approaches provide no support for this challenge. Importantly, service selection differs significantly from product selection, which is the problem addressed by traditional product recommender approaches. The assumptions underlying product recommender approaches do not hold for services. For example a vendor site knows of all product purchases made at it, whereas a service registry does not know of the service episodes that may involve services discovered from it. Also, traditional approaches assume that users are willing to reveal their evaluations to each vendor site. aper formulates the problem of service selection. It reformulates two traditional recommender approaches for service selection and proposes a new agent-based approach in which agents cooperate to evaluate service providers. In this approach, the agents rate each other, and autonomously decide how much to weigh each other’s recommendations. The underlying algorithm with which the agents reason is developed in the context of a concept lattice, which enables finding relevant agents. large service selection datasets do not yet exist, for the purposes of evaluation, we reformulate the well-known product evaluations dataset MovieLens as a services dataset. We use it to compare the various approaches. Despite limiting the flow of information, the proposed approach compares well with the existing approaches in terms of some accuracy metrics defined within.