SIFT Researcher's collaboration with the University of Maryland College Park reported in two papers
SIFT Researcher Ugur Kuter has recently collaborated with Dr. Jennifer Golbeck and her students at the University of Maryland College Park, studying computational models for social trust and social interactions in adversarial scenarios. This work will be reported in two recent papers entitled "CareTaker: A Social Game for Studying Trust Dynamics" and "Coevolving Strategies in Social-Elimination Games." The first paper presents a new Web-based game that provides a framework for studying trust and understanding how trust affects social strategies developed by human players. The second paper provides coevolutionary analysis of how different strategies evolve in a class of social games via an evolutionary formalism for modeling a player that is capable of learning and adapting its strategies to act with the other members of its own group and against those of a conflicting group. Both papers will be presented at the 3rd IEEE International Conference on Social Computing (SocialCom2011).
Paper Information:
Coevolving Strategies in Social-Elimination Games.
K-L. Cheng, U. Kuter, and J. Golbeck. 2011. Proceedings of the 3rd IEEE International Conference on Social Computing (SocialComm?-11). To Appear.
Abstract: Social games provide a framework to study how in- dividuals and groups of individuals interact with each other, and develop strategies to achieve their own well-defined objectives. This paper provides a coevolutionary analysis of how different strategies evolves in a class of social games. We describe a new formalism for modeling a player that is capable of learning and adapting its strategies to act with the other members of its own group and against those of a conflicting group. We define two example player models based on the formalism. We also propose a methodology to compare different player models using coevolution. Our experimental results show that different groups develop, in an alternating fashion, counter- strategies against the other groups over time and the performance a player depends on the player’s ability to adapt to its opponent’s strategies.
Caretaker: A Social Game for Studying Trust Dynamics.
- Violi, J. Golbeck, K-L. Cheng, and U. Kuter. 2011. Proceedings of the 3rd IEEE International Conference on Social Computing (SocialComm?-11). To Appear.
Abstract: We present a new strategy for studying trust deter- mination, especially in the context of dynamic trust. We created a game in which the players start with neutral trust for each other, have assigned roles (“good” or “bad”) and an incentive to deduce the roles of the other players, and are encouraged to take part in social and economic interactions with each other, thereby gathering data with which to make trust determinations. By running user tests, we show that a game with these components generates useful data with which the players’ decisions regarding trust for each other can be observed and better understood. We show that players are more likely to increase their self-reported trust for each other after engaging in an economic transaction, and, surprisingly, their trust for each other increases by a larger margin when the transaction was asymmetric. We also present an analysis of the chat that the players participated in, via an in-game chat system. We show that the two most popular topics of discussion are trade requests and game administration, and suggest possible future work in comparing the players’ chat actions to their trust for each other.