Towards Structure-aware Model for Multi-modal Knowledge Graph Completion

Source code of the paper "Towards Structure-aware Model for Multi-modal Knowledge Graph Completion".This paper was accepted for TMM'2025.

Requirements

• python>=3.9
• torch>=2.0
• transformers
• scipy
• tqdm ...

All experiments are run with 8 V100(32GB) GPUs.

How to run simply

For better reproducibility of the paper, we provide a simple one-stop operation to run the TSAM model.

For DB15K and MKG-W datasets, we use files from DB15K and MKG-W.

MKG-W dataset

Step 1, Please download the tokens folder from Google drive and put it in the TSAM folder. (Due to GitHub storage restrictions, we have stored all processed tokens information in Google drive)

Step 2, Install the model and pre-install related environment

pip install -r requirements.txt

Step 3, Training and evaluate the model

bash train_MKG_W.bash

Feel free to change the output directory to any path you think appropriate.

further

1.If you need it, we also provide the ckpt from our models in the ckpt directory.

2.You can download various transformer-based models from HuggingFace on your own and conduct your own experiments based on the "save_token_embedding.py" py scripts.