Escabilidade do problema de geração de estruturas de coalizão: aplicação de um algoritmo baseado em detecção de comunidades a grafos reais

Abstract

This study presents experimental results on the scalability of the coalition’s structures formation for real graphs that go from 5 thousand vertices. A simple heuristic algorithm called Propagation Algorithm for Coalition Structure Formation (APFEC in Portuguese) with experimental guarantees is presented and probed, based on a balanced label propagation version for detection of communities in very large graphs. The limits of the proposal are evaluated, comparing it with the state-of-the-art in relation to the exact algorithms (ODP-IP - The IP algorithm, based on representation of integer partitions, and ODP, optimal dynamic programming) and heuristic (CFSS - Coalition Formation for Sparse Synergies). The experiments are performed with a set of 14 real-world graphs, and the results show that this approach can calculate coalition structures quickly, even in the presence of the limitations discussed. Finally, the preliminary results are analyzed considering the influence of the ability and the interrelation between the agents in the evaluation of the coalitions.