Probabilistic Trust Models for Social and Service Networks

Trust is a fundamental basis for interaction in multiagent systems. Probabilistic trust models support an agent estimating trustworthiness of another agent based on prior experience. We extend and enhance an existing probabilistic trust model to (1) enable agents to establish trust relations between each other based on no more than limited experience, and (2) provide approaches for agents to find and choose desired interaction partners based on the trust they place in others. Our approaches are inspired by challenges arising in the context of social networks and service-oriented computing. All our approaches can be integrated into the base trust model seamlessly.;When a truster needs to estimate the trust to place in a trustee with whom the truster has little prior experience, the truster must rely upon intermediary agents to lead it to one or more witnesses who would have direct experience with the trustee. Trust propagation approaches that exploit the imperfectly transitive nature of trust address this very problem. Where trustworthy witnesses can be found, such approaches can be quite effective. We propose a generalized way to propagate trust that accommodates situations where trustworthy witnesses cannot be found through typical propagation. Our approach supports an agent customizing the patterns by which to search for witnesses. The resulting approach is generic in that it can capture existing network-based approaches merely through the introduction of appropriate patterns.;In service-oriented computing, we consider service selection wherein service consumers choose services with desired nonfunctional properties. To do this, a service consumer needs trust-based approaches to analyze its service network, which is composed of the trust relations the consumer builds from its interactions with service providers. Existing service selection approaches either fail to capture the dynamic relationships between services or assume that the environment is fully observable. Besides, they do not emphasize the relationship between a consumer's interests and the utility the consumer draws from a service. We propose two trust models that build trust by accurately punishing or rewarding services in terms of the qualities they offer. Both models are effective despite incomplete observations regarding the services under consideration. We also design a service selection approach, built on our trust models, to enable consumers to select services based on their individual utility functions, which reflect their preferences, and estimate trust to capture the providers' quality. This approach guides consumers to construct service compositions that satisfy their quality requirements. An extension of this approach with contracts approximates Pareto optimality without the use of a market mechanism.