Awesome

GEM-GCN

This repository is the official implementation of Generalized Multi-Relational Graph Convolution Network in The Web Conference (WWW) 2021. We follow the code style of tkipf/gcn.

Requirements

python >= 3.6.0
tensorflow = 1.15.0
scipy = 1.4.1

Datasets

The initial datasets of knowledge graph alignment task can be found in JAPE.
The AM dataset can be found in RGCN.
The wordnet dataset can be found in DHNE.
The FB15k dataset can be found in DKRL.

Reproduce Results

1. Entity Alignment

To train models for entity alignment task, run these commands:

./run_align.sh 0 zh_en QuatE --save
./run_align.sh 0 ja_en QuatE --save
./run_align.sh 0 fr_en QuatE --save

where 0 is the GPU index. zh_en is the name of dataset. QuatE is the name of knowledge graph completion method incorporated by our model, feel free to replace it with: RotatE, TransE, TransD, TransH, DistMult.

2. Relation Alignment

To train models for relation alignment task, run these commands:

./run_rel_align.sh 0 zh_en TransE --save
./run_rel_align.sh 0 ja_en TransE --save
./run_rel_align.sh 0 fr_en TransE --save

3. Entity Classification

To train models for entity classification task, run these commands:

./run_class.sh 0 wordnet TransE --save
./run_class.sh 0 fb15k TransE --save
./run_class.sh 0 am TransE --save

For AM dataset, you need to generate am.pickle following RGCN and place it into /data/class. However, we do not recommend using this dataset because its number of labels is very limited.