-
Observation: We show that heuristic sampling schemes limit the generalization ability of deep learning-based hyperedge-prediction.
-
Solution: AHP learns to sample negative examples by adversarial training for better generalization. In terms of AUROC, AHP is up to 28.2% better than best existing methods and up to 5.5% better than variants with sampling schemes tailored to test sets.
-
Experiments: We compare AHP with three sampling schemes and three recent hyperedge-prediction methods on six real hypergraphs
The main paper is available at Here.
| Name | #Nodes | #Edges | Domain |
|---|---|---|---|
| Citeseer | 1,457 | 1,078 | Co-citation |
| Cora | 1,434 | 1,579 | Co-citation |
| Cora-A | 2,388 | 1,072 | Authorship |
| Pubmed | 3,840 | 7,962 | Co-citation |
| DBLP-A | 39,283 | 16,483 | Authorship |
| DBLP | 15,639 | 22,964 | Collaboration |
All datasets are available at Here.
Each dataset file contains following keys: 'N_edges', 'N_nodes', 'NodeEdgePair', 'EdgeNodePair', 'nodewt', 'edgewt', 'node_feat'.
We also provide preprocessed splits, each of which contains train, validation, and test sets (both positive and negative).
They can be found in split/ in the provided link above.
The source code used in the paper is available at ./SIGIR22-AHP/.
python hyperedge_prediction.py --dataset_name cora --model hnhn --epochs 200 --train_DG 1:1
More details about arguments are described in ./SIGIR22-AHP/utils.py.