Source code and datasets for IJCAI-2019 paper "Multi-view Knowledge Graph Embedding for Entity Alignment".
We used two datasets, namely DBP-WD and DBP-YG, which are based on DWY100K proposed in BootEA.
In "data/BootEA_datasets.zip", we give the full data of the two datasets that we used. Each dataset has the following files:
- ent_links: all the entity links without training/test/valid splits;
- 631: entity links with training/test/valid splits, contains three files, namely train_links, test_links and valid_links;
- rel_triples_1: relation triples in the source KG, list of triples like (h \t r \t t);
- rel_triples_2: relation triples in the target KG;
- attr_triples_1: attribute triples in the source KG;
- attr_triples_2: attribute triples in the target KG;
- entity_local_name_1: entity local names in the source KG, list of pairs like (entity \t local_name);
- entity_local_name_2: entity local names in the target KG;
- predicate_local_name_1: predicate local names in the source KG, list of pairs like (predicate \t local_name);
- predicate_local_name_2: predicate local names in the target KG.
The raw datasets of DWY100K can also be found here.
- Python 3
- Tensorflow 1.x
- Numpy
To run the experiments, use:
bash run.sh -m mode -d dataset_folder_path
- mode: training mode, using either ITC or SSL;
- dataset_folder_path: the folder path of dataset to run.
For example, to run the experiments on DBP-WD with ITC mode, use:
bash run.sh -m ITC -d BootEA_DBP_WD_100K/
If you have any difficulty or question in running code or reproducing experimental results, please email to qhzhang.nju@gmail.com, zqsun.nju@gmail.com and whu@nju.edu.cn.
If you use this model or code, please kindly cite it as follows:
@inproceedings{MultiKE,
author = {Qingheng Zhang and Zequn Sun and Wei Hu and Muhao Chen and Lingbing Guo and Yuzhong Qu},
title = {Multi-view Knowledge Graph Embedding for Entity Alignment},
booktitle = {IJCAI},
pages = {5429--5435},
year = {2019}
}