ChapTER

Deja vu: Contrastive Historical Modeling with Prefix-tuning for Temporal Knowledge Graph Reasoning

Official code repository :bookmark_tabs: for NAACL 2024 Findings paper "Deja vu: Contrastive Historical Modeling with Prefix-tuning for Temporal Knowledge Graph Reasoning".

:link: Arxiv, ACL Anthology

Requirements

• python>=3.7
• torch>=1.6
• transformers>=4.15

All experiments are run with GeForce RTX 3090 GPUs.

How to Run

We provide codes and instructions for running transductive TKG reasoning experiments.

Total 3 steps to follow: dataset preprocessing, model training, and model evaluation.

Datasets used in this repository are list in the directory $REPO_DIR/data. Besides, we've preprocessed the datasets, so you can skip step 1 (unless you want to preprocess your own datasets), directly run the training and evaluation scripts.

Running Steps

Taking ICEWS14 dataset as an example:

Step 1, preprocess the dataset

bash scripts/preprocess.sh ICEWS14

Step 2, training the model

CUDA_VISIBLE_DEVICES=0,1,2 OUTPUT_DIR=./checkpoint/ICEWS14/ bash scripts/train_icews14.sh

Step 3, evaluate the model

CUDA_VISIBLE_DEVICES=0 bash scripts/eval.sh ./checkpoint/ICEWS14/model_last.mdl ICEWS14

The trained model checkpoints and output files are saved in directory $REPO_DIR/checkpoint.

Note: For custom dataset, replace ICEWS14 to any dataset name that you want to run.

Citation

If you use our code in your research, please cite our work:

@inproceedings{peng2024deja,
  title={Deja vu: Contrastive Historical Modeling with Prefix-tuning for Temporal Knowledge Graph Reasoning},
  author={Peng, Miao and Liu, Ben and Xu, Wenjie and Jiang, Zihao and Zhu, Jiahui and Peng, Min},
  booktitle={Findings of the Association for Computational Linguistics: NAACL 2024},
  pages={1178--1191},
  year={2024}
}