Computational Trust Model for Repeated Trust Games

Abstract

Trust game is a money exchange game that has been widely used in behavioral economics for studying trust and collaboration between humans. In this game, exchange of money is entirely attributable to the existence of trust between users. The trust game could be one-shot, i.e. the game ends after one round of money exchange, or repeated, i.e. it lasts several rounds. Predicting user behavior in the repeated trust game is of critical importance for the next movement of the partners. However, existing behavior prediction approaches uniquely rely on players personal information such as their age, gender and income and do not consider their past behavior in the game. In this paper, we propose a computational trust metric that is uniquely based on users past behavior and can predict the future behavior in repeated trust game. Our trust metric can distinguish between users having different behavioral profiles and is resistant to fluctuating user behavior. We validate our model by using an empirical approach against data sets collected from several trust game experiments. We show that our model is consistent with rating opinions of users, and our model can provide higher accuracy on predicting users' behavior compared with other naive models.