A novel game playing based approach to the modeling and support of consensus reaching in a group of agents

The paper concerns the problem of reaching consensus among agents in group decision making. A popular framework of individual preferences expressed as (fuzzy) preference relations is adopted. The consensus reaching process is assumed to be based on a discussion in the group of agents, which is expected to make the initially expressed preferences closer one to another. We present a novel approach to the modeling of the consensus reaching process as a game, in the sense of game playing. We use the Monte Carlo Tree Search (MCTS) algorithm with the Upper Confidence Bounds Applied for Trees selection formula, which is a state of the art solution algorithm in that area. The consensus reaching process is modeled as a sequence of actions, referred to as moves, of the individual agents involved. A model of the assessment of a configuration of the individual preference relations is proposed. A decision support system that implements the approach proposed is developed, which provides the agents with an easy to read evaluation of the expected outcome of each move. The approach constitutes a new paradigm in the modeling of a consensus reaching process, and then its support.