Coalition Formation: Concessions, Task Relationships and Complexity Reduction

Abstract

Solutions to the coalition formation problem commonly assume agent rationality and, correspondingly, utility maximization. This in turn may prevent agents from making compromises. As shown in recent studies, compromise may facilitate coalition formation and increase agent utilities. In this study we leverage on those new results. We devise a novel coalition formation mechanism that enhances compromise. Our mechanism can utilize information on task dependencies to reduce formation complexity. Further, it works well with both cardinal and ordinal task values. Via experiments we show that the use of the suggested compromise-based coalition formation mechanism provides significant savings in the computation and communication complexity of coalition formation. Our results also show that when information on task dependencies is used, the complexity of coalition formation is further reduced. We demonstrate successful use of the mechanism for collaborative information filtering, where agents combine linguistic rules to analyze documents' contents.