Relational clustering starting from non-Euclidean dissimilarity relations