Decentralized Multiagent Approach for Hedonic Games

We propose a novel, multi-agent, decentralized approach for hedonic coalition formation games useful for settings with a large number of agents. We also propose three heuristics which, can be coupled with our approach to find sub-coalitions that prefer to "bud off" from an existing coalition. We found that our approach when compared to random partition formation gives better results which further improve when it is coupled with the proposed heuristics. As matching problems are a common type of hedonic games, we have adapted our approach for two matching problems: roommate matching and bipartite matching. Our method does well for additively separable hedonic games, where finding the optimal partition is NP-hard, and gives near optimal results for matching problems.