Unequal Singleton Pair Distance for Evidential Preference Clustering.

Evidential preference based on belief function theory has been proposed recently, simultaneously characterizing preference information with uncertainty and imprecision. However, traditional distances on belief functions do not adapt to some intrinsic properties of preference relations, especially when indifference relation is taken into comparison, therefore may cause inconsistent results in preference-based applications. In order to solve this issue, Unequal Singleton Pair (USP) distance has been proposed previously, with applications limited in preference aggregation. This paper explores forward the effectiveness of USP distance in preference clustering, especially confronting multiple conflicting sources. Moreover, a combination strategy for multiple conflicting sources of preference is proposed. The experiments on synthetic data show that USP distance can effectively improve the clustering results in Adjusted Rand Index (ARI).