Genetic Approach to Stable Partitioning in Online Role Based Hedonic Games

Abstract

In role-based hedonic games, agents are partitioned into teams and matched to suitable roles within their teams. A stable partition and matching is one in which each agent accepts its role and the composition of roles fulfilled on the team. To achieve this, a population must have a suitable distribution of preferences for roles and compositions such that supply and demand can be fulfilled. Many NP-complete problems arise in attempting to match agents into acceptable teams. We propose a genetic local search approach to matching players into stable teams. Our approach adapts to changes in preferences over time, and we validate our approach on real world matchmaking data from an online game, League of Legends. We show improvements on three optimization criteria over greedy local search, and finally, we show how updates to the chromosome vector can be interpreted to discover deficiencies in the supply and demand of particular roles.