helios_code's Introduction

HELIOS (Hyper-relational Schema Model)

HELIOS is a hyper-relational schema model, which directly learns from hyper-relational schema tuples in a KG. HELIOS captures not only the correlation between multiple types of a single entity, but also the correlation between types of different entities and relations in a hyper-relational schema tuple. Please see the details in our paper below:

Yuhuan Lu, Bangchao Deng, Weijian Yu, and Dingqi Yang. 2023. HELIOS: Hyper-Relational Schema Modeling from Knowledge Graphs. In Proceedings of the 31st ACM International Conference on Multimedia (MM ’23), October 29–November 3, 2023, Ottawa, ON, Canada.

How to run the code

Train and evaluate model (suggested parameters for JF17k, WikiPeople and WD50K dataset)

python run.py --dataset jf17k --gpu 0

python run.py --dataset wikipeople --gpu 0

python run.py --dataset wd50k --gpu 0

The datasets are available here: https://www.dropbox.com/s/iz5wxp0uldx5i05/data.zip?dl=0 , and put them into the data folder.

Parameter setting:

In run.py, you can set:

--dataset: input dataset

--epochs: number of training epochs

--batch_size: batch size of training set

--dim: embedding size

--learning_rate: learning rate

--self_attention_layers: number of self-attention layers

--gat_layers: number of GAT layers

--gpu: gpu to be used for train and test the model

--num_attention_heads: number of attention heads

Python lib versions

Python: 3.7.13

torch: 1.11.0

Reference

If you use our code or datasets, please cite:

@inproceedings{lu2023helios,
  title={HELIOS: Hyper-Relational Schema Modeling from Knowledge Graphs},
  author={Lu, Yuhuan and Deng, Bangchao and Yu, Weijian and Yang, Dingqi},
  booktitle={Proceedings of the 31st ACM International Conference on Multimedia},
  pages={4053--4064},
  year={2023}
}

Recommend Projects