A geometrical method for consensus building in GDM with incomplete heterogeneous preference information