Some questions about ground truth similarity in the paper

Hi, sorry to disturb you. I found that you still used the ground truth score in the training process. However, I wonder if I have graphs with no ground truth labels, is it possible for me to use a similar approach to compute the similarity between g1 and g2? Thanks.