This is the source code of our paper Each Snapshot to Each Space: Space Adaptation for Temporal Knowledge Graph Completion published in ISWC2022.
- The links of the original datasets ICEWS14 and ICEWS05-15 can be found from paper: Learning Sequence Encoder for Temporal Knowledge Graph Completion.
- YAGO11k and Wikidata12k have been preprocessed with reference to Temporal Knowledge Graph Completion Based on Time Series Gaussian Embedding.
-
model.py contains PTorch(1.x) based implementation of our proposed models.
-
To reproduce the reported results of our model, use the following commands:
python main.py --model SANe --name sane --lr 0.001 --data icews14 --train_strategy one_to_x python main.py --model SANe --name sane --lr 0.001 --data icews05-15 --k_h 30 --embed_dim 300 --feat_drop 0.2 --hid_drop 0.3 --ker_sz 7 --train_strategy one_to_x --batch 512 python main.py --model SANe --name sane --lr 0.001 --data yago --ker_sz 5 --train_strategy one_to_x python main.py --model SANe --name sane --lr 0.001 --data wikidata --ker_sz 5 --train_strategy one_to_x