Abstract

Social networking sites have developed considerably during the past couple of years. However, few websites exploit the potentials of combining the social networking sites with online markets. This, in turn, would help users to distinguish and engage into interaction with other unknown, yet trustworthy, users in the market. In this thesis, we develop a model to estimate the trust of unknown agents in a multi-agent system where agents engage into business-oriented interactions with each other. The proposed trust model estimates the degree of trustworthiness of an unknown target agent through the information acquired from a group of advisor agents, who had direct interactions with the target agent. This problem is addressed when: (1) the trust of both advisor and target agents is subject to some uncertainty; (2) the advisor agents are self-interested and provide misleading accounts of their past experiences with the target agents; and (3) the outcome of each interaction between the agents is multi-valued.

We use possibility distributions to model trust with respect to its uncertainties thanks to its potential capability of modeling uncertainty arisen from both variability and ignorance. Moreover, we propose trust estimation models to approximate the degree of trustworthiness of an unknown target agent in the two following problems: (1) in the first problem, the advisor agents are assumed to be unknown and have an unknown level of trustworthiness; and (2) in the second problem, however, some interactions are carried out with the advisor agents and their trust distributions are modeled. In addition, a certainty metric is proposed in the possibilistic domain, measuring the
confidence of an agent in the reports of its advisors which considers the consistency in the advisors’ reported information and the advisors’ degree of trustworthiness. Finally, we validate the proposed approaches through extensive experiments in various settings.