MRME-KGC

Source code of the paper "Multi-view Riemannian Manifolds Fusion Enhancement for Knowledge Graph Completion".This paper was submitted to TKDE'2024

Requirements

pip install -r requirement.txt

Details such like:

• python>=3.8
• torch>=1.8
• tqdm
• geoopt
• sklearn

All experiments are run with 4 RTX3090(24GB) GPUs.

Data Preparation

1. Due to the size limit of github, please download the dataset (FB15K-237, CN-100K, Kinships, WN18RR, YAGO3-10, UML.) from google drive and save it in the following format:

• MRME-KGC
  • model
    • manifolds
      • file1
      • file2
    • file3 .....
  • src_data
    • CN-100K
    • FB237
    • Kinships
    • UML
    • WN18RR
    • YAGO3-10

2.After completing the above steps, please run the "process_datasets.py" file in the model folder to generate KG data to be input into the model:

cd model/
python process_datasets.py

How to Run

FB15K-237 dataset

bash Run-FB237-100d.sh

CN100K dataset

bash Run-CN-100K-100d.sh

UML dataset

bash Run-UML-100d.sh

....

Troubleshooting

1. I encountered "CUDA out of memory" when running the code.

We run experiments with 4 RTX3090(24GB) GPUs, please reduce the batch size if you don't have enough resources.

2. ModuleNotFoundError: No module named 'sklearn'

Please note that the command used to install the 'sklearn' package is: "pip install -U scikit-learn" instead of "pip install -U sklearn". Please pay attention to this problem!

Drawing

If you want to draw pictures similar to the model pictures in the paper about hyperbolas and spheres, please refer to the "Draw a picture of hyperbolic space.ipynb" file.

vim Draw a picture of hyperbolic space.ipynb

🤝 Connection

If you have any questions, please contact my email [email protected] !