Game theoretic Markov decision processes for optimal decision making in social systems

One key problem in social systems is to understand how users learn and make decision. Since the values of social systems are created by user participation while the user-generated data is the outcome of users' decisions, actions and their social-economic interactions, it is very important to take into account users' local behaviors and interests when analyzing a social system. In this paper, we propose a game-theoretic Markov decision process (GTMDP) framework to study how users make optimal decisions in a social system. By explicitly considering users' local interactions and interests, we show that the proposed GTMDP can correctly derive the optimal decision and thus achieve much better expected long-term utility compared with the traditional MDP. We also discuss how to design mechanism to steer users' behavior under the proposed GTMDP framework.