AutoETER

EMNLP 2020 Findings: AutoETER: Automated Entity Type Representation for Knowledge Graph Embedding Paper.

This is our source code and data for the paper:

Guanglin Niu, Bo Li, Yongfei Zhang, Shiliang Pu and Jingyang Li. AutoETER: Automated Entity Type Representation for Knowledge Graph Embedding. EMNLP 2020 Findings, 2020, 1172–1181.

Author: Dr. Guanglin Niu (beihangngl at buaa.edu.cn)

Introduction

To explore the type information for any KG, we develop a novel KGE framework with Automated Entity TypE Representation (AutoETER), which learns the latent type embedding of each entity by regarding each relation as a translation operation between the types of two entities with a relation-aware attention mechanism. Particularly, our approach could model and infer all the relation patterns and complex relations.

Dataset

We provide four datasets: FB15K, FB15K237, WN18 and YAGO3-10. You can find and download all the datasets on Onedrive.

Example to run the codes

Train and Test on FB15k:

python codes/run.py --cuda --do_train --do_valid --do_test --data_path/FB15k --model AutoETER -n 128 -b 1024 -d 1000  -td 200 -g 22.0 -gt 7.0 -gp 5.0 -al1 0.1 -al2 0.5 -a 1.0 -lr 0.0001 --max_steps 150000 -save models/AutoETER_fb15k_1 -ps 16

Acknowledge

@inproceedings{Niu:AutoETER,
  author    = {Guanglin Niu and
               Bo Li and
               Yongfei Zhang and
               Shiliang Pu and
               Jingyang Li},
  title     = {AutoETER: Automated Entity Type Representation for Knowledge Graph Embedding},
  booktitle = {Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: Findings},
  year      = {2020}
}

yeongjoonJu / AutoETER