This is the code of paper Rethinking Graph Convolutional Networks in Knowledge Graph Completion. Zhanqiu Zhang, Jie Wang, Jieping Ye, Feng Wu. WWW 2022. [arXiv]
- python 3.7
- torch 1.8
- dgl 0.7
Pleaes run the commands in RGCN+CompGCN+LTE/script or WGCN/script to reproduce the results.
Meaning of different options.
- rat: random adjacency tensors.
- wsi: without self-loop information.
- wni: without neighbor information.
- ss: sample set sizes for random sampled neighbors.
If you find this code useful, please consider citing the following paper.
@inproceedings{WWW22_GCN4KGC,
author = {Zhanqiu Zhang and Jie Wang and Jieping Ye and Feng Wu},
booktitle = {The Web Conference 2022},
title = {Rethinking Graph Convolutional Networks in Knowledge Graph Completion},
year = {2022}
}
We refer to the code of CompGCN, WGCN, and DGL. Thanks for their contributions.