GoldE

Generalizing Knowledge Graph Embedding with Universal Orthogonal Parameterization

(This paper is accepted by ICML 2024)

Requirements

• pytorch == 1.8.0
• numpy == 1.19.2
• scikit-learn == 0.23.2

Data

• entities.dict: a dictionary map entities to unique ids
• relations.dict: a dictionary map relations to unique ids
• train.txt: the KGE model is trained to fit this data set
• valid.txt: create a blank file if no validation data is available
• test.txt: the KGE model is evaluated on this data set

Usage

All training commands are listed in best_config.sh. For example, you can run the following commands to train GoldE on WN18RR datasets.

# WN18RR
bash run.sh GoldE wn18rr 0 0 0 1000 200 800 12 10 0.666435178264418 0.99 0.5 6.0 1.1 0.003 60000 20000 16 0.185933138885153 -sf

Acknowledgement

We refer to the code of RotatE and HousE. Thanks for their contributions.