Do Higher-Order Models Win for Higher-Order Reasons? Rethinking Performance Gains in Hypergraph Learning
The paper argues that hypergraph models often keep their edge even when the higher-order signal is deliberately disrupted.
The authors test 25 hypergraph learning benchmarks with a framework that perturbs higher-order information while preserving pairwise structure. They find that higher-order models frequently still outperform lower-order baselines after that perturbation. The remaining gaps shrink when lower-order baselines get stronger pairwise weighting, more feature propagation, and normalization. The paper’s claim is that benchmark wins should not be treated as proof that higher-order information caused the gain. ArXiv · AI/CL/LG's note
The authors test 25 hypergraph learning benchmarks with a framework that perturbs higher-order information while preserving pairwise structure. They find that higher-order models frequently still outperform lower-order baselines after that perturbation. The remaining gaps shrink when lower-order baselines get stronger pairwise weighting, more feature propagation, and normalization. The paper’s claim is that benchmark wins should not be treated as proof that higher-order information caused the gain. ArXiv · AI/CL/LG's note
score 4