Explaining Human Comparisons using Alignment-Importance Heatmaps