Coalition Generation Scalability through Community Detection: Results on Real Graphs

Fabio Sebastian Silveira, Gustavo Alberto Giménez Lugo

Abstract: This article presents experimental results on the scalability of coalition formation for real graphs approaching up to 5000 vertices. A simple heuristic algorithm named Propagation Algorithm for Coalition Structure Formation with no warranties is presented and probed, based on a version of label propagation for community detection in very large graphs [1]. The limits of the proposal are gauged, comparing it with the state-ofthe-art regarding exact (ODP-IP [2]) and heuristic (CFSS [3]) algorithms. Experiments were run over a set of 14 real world, power law graphs, and the results show a more than satisfactory performance, even in the presence of limitations that are discussed. Finally, preliminary results are explored considering the influence of the skill and the inter-relationship between agents in the evaluation of the coalitions.