HypeTKG

This is the code for our EMNLP 2024 paper: Temporal Fact Reasoning over Hyper-Relational Knowledge Graphs (https://arxiv.org/abs/2307.10219).

Installation

Create a conda environment

conda create -n HypeTKG python=3.8
conda activate HypeTKG

Configure HypeTKG requirements

Install PyTorch (specify your own CUDA version)

conda install pytorch==1.12.0 torchvision==0.13.0 torchaudio==0.12.0 cudatoolkit=10.2 -c pytorch

Install PyTorch Geometric: conda install pyg -c pyg

Install PyTorch Scatter: conda install pytorch-scatter -c pyg

Install wandb: conda install -c conda-forge wandb

install tqdm: conda install -c conda-forge tqdm

install numpy: conda install numpy=1.19

Datasets

The dataset can be found in data/Wiki-hy and data/YAGO-hy.

Their derivatives can be found there as well:

• Wiki-hy(33) and YAGO-hy(33) - approx 33% of statements have qualifiers
• Wiki-hy(66) and YAGO-hy(66) - approx 66% of statements have qualifiers
• Wiki-hy(100) and YAGO-hy(100) - 100% of statements have qualifiers

Running Experiments

Parameters

Parameters are available in the CONFIG dictionary in the run.py.

• DATASET: YAGO-hy or Wiki-hy
• SAMPLER_W_QUALIFIERS: True for hyper-relational models [default], False for quadruple-based model only
• SAMPLER_W_STATICS:True for the model with time-invariant facts, False for Hyper-Relational Temporal Knowledge Graphs model only

Training Model

python run.py