PRASE-Python

The code of paper Unsupervised Knowledge Graph Alignment by Probabilistic Reasoning and Semantic Embedding [arxiv] [ijcai] in Proceedings of IJCAI 2021.

Package

The packages of PRASE-Python is presented below.

  ├─PRASE-Python
  │  ├─data: necessary data for running the test script
  │  ├─model: the implementation of PARIS
  │  ├─objects: Entity/Relation/KG/KGs objects of PRASE
  │  ├─test.py: a test script example

Installation

Dependencies

• Python 3.x
• Numpy

Start

Use the following command to get a quick start:

python test.py

This test script performs PRASE on D-W-15K-V2 with both the embedding and the mapping feedback from SE module (i.e., BootEA). You can revise the code to customize the PRASE model.

Usage

The core of this package is actually the Python-version implementation of PARIS. Since this package does not contain the implementations of embedding-based approaches, you should adopt the embedding-based implementations from external libraries to run PRASE, such as OpenEA.

Citation

If you use this model or code, please cite it as follows:

@inproceedings{ijcai2021-278,
  title     = {Unsupervised Knowledge Graph Alignment by Probabilistic Reasoning and Semantic Embedding},
  author    = {Qi, Zhiyuan and Zhang, Ziheng and Chen, Jiaoyan and Chen, Xi and Xiang, Yuejia and Zhang, Ningyu and Zheng, Yefeng},
  booktitle = {Proceedings of the Thirtieth International Joint Conference on
               Artificial Intelligence, {IJCAI-21}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Zhi-Hua Zhou},
  pages     = {2019--2025},
  year      = {2021},
  month     = {8},
  note      = {Main Track}
  doi       = {10.24963/ijcai.2021/278},
  url       = {https://doi.org/10.24963/ijcai.2021/278},
}