Inferring object rankings based on noisy pairwise comparisons from multiple annotators