Revisiting Condorcet Fusion

Abstract

The fusion task is to aggregate ranked document lists retrieved for a query. The Condorcet voting criterion served as inspiration for a commonly used fusion method proposed by Montague and Aslam (2002). The method is stochastic as it is based on the QuickSort sorting algorithm. We empirically show that the performance of the method can substantially vary due to this stochastic aspect. We propose approaches that improve the performance robustness of this fusion method with respect to its stochastic nature. The resultant performance is on par with the state-of-the-art.