The implementation of our work "Attention Enhanced Knowledge Graph Embeddings with Variable Receptive Fields for Link Prediction".
python==3.8
numpy==1.21.5
scikit-learn==1.0.2
scipy==1.7.3
torch==1.12.1
tqdm
We conduct experiments on 7 datasets:
| Datasets | Entities | Relations | Train | Valid | Test |
|---|---|---|---|---|---|
| WN18RR | 40,943 | 11 | 86,835 | 3,034 | 3,134 |
| FB15k-237 | 14,541 | 237 | 272,115 | 17,535 | 20,466 |
| WN18 | 40,943 | 18 | 141,442 | 5,000 | 5,000 |
| FB15k | 14,951 | 1345 | 483,142 | 50,000 | 59,071 |
| KINSHIP | 104 | 25 | 8,544 | 1,068 | 1,074 |
| YAGO3-10 | 123,182 | 37 | 1,079,040 | 5,000 | 5,000 |
Take WN18RR as a example:
python main.py --data_path "./data" --run_folder "./" --data_name "WN18RR" --embedding_dim 200 --filter1_size 1 3 --filter2_size 3 3 --filter3_size 1 5 --output_channel 5 --min_lr 0.001 --batch_size 1024 --log_epoch 2 --neg_ratio 1 --input_drop 0.2 --hidden_drop 0.1 --feature_map_drop 0.1 --opt "Adam" --learning_rate 0.001 --weight_decay 5e-4 --factor 0.5 --verbose 1 --patience 5 --max_mrr 0 --epoch 600 --momentum 0.9 --save_name "./model/wn18rr