This DGL example implements the link prediction model proposed in the paper
Link Prediction Based on Graph Neural Networks
and REVISITING GRAPH NEURAL NETWORKS FOR LINK PREDICTION
The author's codes of implementation is in SEAL (pytorch)
and SEAL_ogb (torch_geometric)
This example was implemented by Smile during his intern work at the AWS Shanghai AI Lab.
ogbl-collab
- NumNodes: 235868
- NumEdges: 2358104
- NumNodeFeats: 128
- NumEdgeWeights: 1
- NumValidEdges: 160084
- NumTestEdges: 146329
- python 3.6+
- Pytorch 1.5.0+
- ogb
- pandas
- tqdm
- scipy
In the seal_dgl folder
run on cpu:
python main.py --gpu_id=-1 --subsample_ratio=0.1run on gpu:
python main.py --gpu_id=0 --subsample_ratio=0.1experiment on ogbl-collab
| method | valid-hits@50 | test-hits@50 |
|---|---|---|
| paper | 63.89(0.49) | 53.71(0.47) |
| ours | 63.56(0.71) | 53.61(0.78) |
Note: We only perform 5 trails in the experiment.