A comparative analysis and improvement of MaxSAT encodings for coalition structure generation under MC-nets

Abstract

Coalition structure generation (CSG) is one of the main research issues in the use of coalitional games in multiagent systems and weighted partial MaxSAT (WPM) encodings, i.e. rule relation-based WPM (RWPM) and agent relation-based WPM (AWPM), which are efficient for solving the CSG problem. Existing studies show that AWPM surpasses RWPM since it achieves more compact encoding; it generates fewer variables and clauses than RWPM. However, in this paper, we focus on a special case in which the two encodings generate identical numbers of variables and clauses. Experiments show that RWPM surprisingly has a dominant advantage over AWPM, which aroused our interest. We exploit the deep-rooted reason and find that it is the redundancy when encoding transitive laws in RWPM that leads to this situation. Finally, we remove redundant clauses for transitive laws in RWPM and develop an improved RWPM with refined transitive laws to solve the CSG problem. Experiments demonstrate that refined encoding is more compact and efficient than previous WPM encodings.