IncDE

The codes and datasets for "Towards Continual Knowledge Graph Embedding via Incremental Distillation" [AAAI 2024].

Framework

Folder Structure

The structure of the folder is shown below:

 IncDE
 ├─checkpoint
 ├─data
 ├─logs
 ├─save
 ├─src
 ├─main.py
 ├─data_preprocess.py
 └README.md

Introduction to the structure of the folder:

/checkpoint: The generated models are stored in this folder.
/data: The datasets(ENTITY, RELATION, FACT, HYBRID, graph_equal, graph_higher, graph_lower) are stored in this folder.
/logs: Logs for the training are stored in this folder.
/save: Some temp results are in this folder.
/src: Source codes are in this folder.
/main.py: To run the IncDE.
data_preprocess.py: To prepare the data processing.
README.md: Instruct on how to realize IncDE.

Requirements

All experiments are implemented on the NVIDIA RTX 3090Ti GPU with the PyTorch. The version of Python is 3.7.

Please run as follows to install all the dependencies:

pip3 install -r requirements.txt

Usage

Preparation

Unzip the dataset $data1.zip$ and $data2.zip$ in the folder of $data$.
Prepare the data processing in the shell:

python data_preprocess.py

Main Results

Run the code with this in the shell:

python main.py -dataset ENTITY -gpu 0

Ablation Results

Run the code with this in the shell:

./ablation.sh

Citation

If you find this method or code useful, please cite

@inproceedings{liu2024towards,
  title={Towards Continual Knowledge Graph Embedding via Incremental Distillation},
  author={Liu, Jiajun and Ke, Wenjun and Wang, Peng and Shang, Ziyu and Gao, Jinhua and Li, Guozheng and Ji, Ke and Liu, Yanhe},
  booktitle={AAAI},
  year={2024}
}

seukgcode/IncDE