NeurIPS 2020

Learning to Extrapolate Knowledge: Transductive Few-shot Out-of-Graph Link Prediction


Meta Review

This paper consider few-shot link prediction in the "out of graph" setting, in which novel nodes that were not available at training time are introduced. The reviewers initially had questions about the novelty of the approach and the baselines considered in the experiments. These questions were sufficiently addressed in the authors' response, which clarified that the emphasis on the out of graph setting was the main novelty and provided additional experiments to show that other baselines did not handle this problem as well. The reviewers strongly encourage the reviewers to add these results and clarifications to the camera ready version of the paper. After discussion, they agreed that the paper was ready for publication and made a novel contribution to an interesting problem that has not received much attention. We note that Reviewer 2 asks some follow up questions about the new results and the authors are also strongly encouraged to address them.