The paper proposes to generalize message passing neural networks (MPNN) for modeling multi-relational, recursive and ordered hypergraphs, which is theoretically grounded and challenging topic. The authors tested the proposed framework on semi-supervised node classification and link prediction tasks on several (7) benchmark datasets, where the proposed methods outperformed other baselines consistently. Although some reviewers point out the presentation issues with unclear notations and suggest ways to improve and more baselines in addition, the merits of the paper outweigh the drawbacks and acceptance is recommended. We strongly encourage the authors to take the reviewers’ feedback into account in their revision of the paper.