We would like to thank all reviewers for your helpful suggestions

We would like to thank all reviewers for your helpful suggestions. In what follows we answer your questions to help clarify the notations and experiments. "motivation not clear": G-SSL and specifically label propagation used to be an important semi-supervised learning We will try to include more real-world examples in the revised version. ": following the standard notations, we use The feature matrix is composed of labeled part and unlabeled part. "Datasets not explained": we will add more descriptions of datasets in the next version.