Comparison-based centrality measures

Abstract Recently, learning only from ordinal information of the type “item x is closer to item y than z ” has received increasing attention in machine community. Such triplet comparisons are particularly well suited for crowdsourced human intelligence tasks, which workers make statements about relative distances a items. In this paper, we systematically investigate comparison-based centrality measures on triplets and theoretically analyze their underlying Euclidean notion centrality. Two such already appear literature under opposing approaches, propose third measure, natural compromise between these two. We further discuss relation statistical depth functions, comprise desirable properties measures, conclude with experiments real synthetic datasets medoid estimation outlier detection.