PAAR

This repository contains our implementation of paper Multi-hop Knowledge Graph Reasoning Based on Hyperbolic Knowledge Graph Embedding and Reinforcement Learning Xingchen Zhou, Peng Wang, Qiqing Luo, Zhe Pan

1. Dependencies

• Python 3.6+
• Pytorch 1.0+

2. Results

The results of PAAR on FB15k-237-20% and NELL23K are shown as follows.

Datasets	MRR	H@13	H@10
FB15k-237-20%	.241	.263	.375
NELL23K	.184	.202	.309

3. Instructions for running

To analyze PAAR's performance on the datasets, please run our code as follows.

3.1 Data Process

./experiment.sh configs/<dataset>.sh --process_data <gpu-ID>

3.2 Hyperbolic Knowledge Graph Embedding

./experiment-emb.sh configs/<dataset>-conve.sh --train <gpu-ID>

3.3 Train

./experiment-rs.sh configs/<dataset>-rs.sh --train <gpu-ID> 

3.4 Test

./experiment-rs.sh configs/<dataset>-rs.sh --inference <gpu-ID> 

4. Acknowledgement

We refer to the code of DacKGR. Thanks for their contributions.

5. Licences

Every source code file written exclusively by the author of this repo is licensed under Apache License Version 2.0. For more information, please refer to the license.