Weimplement our approach using TensorFlow[1]forefficient GPU-based computation and also use some components of TensorFlow Probability [4]2.

Graph neural networks (GNNs) have shown the power in graph representation learningfornumeroustasks.

Graph contrastivelearning attracts/disperses node representations forsimilar/dissimilar node pairs under some notion of similarity.

In Figure 1 we demonstrate the common neighbor (CN) distribution among positive and negative test samples for ogbl-collab, ogbl-ppa, and ogbl-citation2. These results demonstrate that a vast majority of negative samples have no CNs. Since CNs is a typically good heuristic, this makes it easy to identify most negative samples. We further present the CN distribution of Cora, Citeseer, Pubmed, and ogbl-ddi in Figure 3. The CN distribution of Cora, Citeseer, and Pubmed are consistent with our previous observations on the OGB datasets in Figure 1. We note that ogbl-ddi exhibits a different distribution with other datasets. As compared to the other datasets, most of the negative samples in ogbl-ddi have common neighbors. This is likely because ogbl-ddi is considerably denser than the other graphs.

Link prediction attempts to predict whether an unseen edge exists based on only a portion of edges of a graph. A flurry of methods have been introduced in recent years that attempt to make use of graph neural networks (GNNs) for this task. Furthermore, new and diverse datasets have also been created to better evaluate the effectiveness of these new models. However, multiple pitfalls currently exist that hinder our ability to properly evaluate these new methods. These pitfalls mainly include: (1) Lower than actual performance on multiple baselines, (2) A lack of a unified data split and evaluation metric on some datasets, and (3) An unrealistic evaluation setting that uses easy negative samples. To overcome these challenges, we first conduct a fair comparison across prominent methods and datasets, utilizing the same dataset and hyperparameter search settings. We then create a more practical evaluation setting based on a Heuristic R elated Sampling Technique (HeaRT), which samples hard negative samples via multiple heuristics. The new evaluation setting helps promote new challenges and opportunities in link prediction by aligning the evaluation with real-world situations.

Following this trend, SIG-6 VAE is estimated to take at least 600GRAM on Cora with their original setting7 (K = 150,J = 2000),which isanormally unaffordable memory. We note a full matrix forU32 could provide extra flexibility to model stochastic equivalence/disassortativity (e.g., in protein-protein interaction33 network), while a diagonal one is more suitable to model an assortative relational network exhibiting homophily34 (e.g., co-author network)butnotnecessarily stochasticequivalence.

The results are shown in Figure 6.

We thank all the reviewers for their constructive feedback. Reviewer #1: (1) Number of labeled nodes to train the policy network. ANRMAB, at least a moderate number of labeled data are required. We observe similar trends to the results in Section 4.4 (Paper). We have compared classification performance w.r.t.