An agent-based service marketplace for dynamic and unreliable settings

Abstract

In order to address the unreliable nature of service providers, and the dynamic nature of services (their quality values could change frequently over time due to various factors), this paper proposes a probabilistic, multi-valued quality model for services, capable of capturing uncertainty in their quality values by assigning each quality attribute with multiple potential values (or ranges of values), along with a corresponding probability distribution over these values. The probability distribution indicates the most likely quality value for an attribute at the current time step, but also notifies discovery applications of the possibility of other, possibly worse outcomes, thus ultimately facilitating more reliable service selection and composition via avoiding services with high uncertainty. Such uncertainty-aware, multi-valued quality models of services are maintained via an agent-based service marketplace, where each service is associated with a software agent, capable of learning the time-varying probability distributions of its quality values through applying online learning techniques, based on the service’s past performance information. The experiments conducted demonstrate the effectiveness of the proposed approach.