A trust model with statistical foundation

Abstract

The widespread use of the Internet signals the need for a better understanding of trust as a basis for secure on-line interaction. In the face of increasing uncertainty and risk, users and machines must be allowed to reason effectively about the trustworthiness of other entities. In this thesis, we propose a trust model that assists users and machines with decision-making in online interactions by using past behavior as a predictor of likely future behavior. We develop a general method to automatically compute trust based on self-experience and the recommendations of others. Our trust model solves the problem of recommendation combination and detection of unfair recommendations. Our approach involves data analysis methods (Bayesian estimation, Dirichlet distribution), and machine learning methods (Weighted Majority Algorithm). Furthermore, we apply our trust model to several utility models to increase the accuracy of decision-making in different contexts of Web Services. We describe simulation experiments to illustrate its effectiveness, robustness and the evolution of trust.