Source code for LargeEA: Aligning Entities for Large-scale Knowledge Graphs

If there is any problem with reproduction, please create an issue in The GitHub repo page

pytorch>=1.7.0

tensorflow>=2.4.1 (required for RREA)

faiss

transformers

datasketch[redis]

...

A full list of required packages is located in src/requirements.txt

The IDS benchmark is provided by OpenEA

Our newly proposed benchmark DBP1M is available at Google Drive

First download and unzip dataset files, place them to the project root folder:

unzip OpenEA_dataset_v1.1.zip
unzip mkdata.zip

The dataset (small for IDS15K, medium for IDS100K, large for DBP1M) and lang (fr or de) parameter controls which benchmark to use. For example, in the src folder, setting dataset to small and lang to fr will run on OpenEA EN_FR_15K_V1 dataset.

Take DBP1M(EN-FR) as an example:

Make sure the folder for results is created:

cd src/
mkdir tmp4

First get the BERT embeddings of all entities

python main.py --phase 1 --dataset large --lang fr 

Then calculate TopK sims based on BERT:

python main.py --phase 2 --dataset large --lang fr 

Finally the string-based similarity(this requires a redis server listening localhost:6379):

python main.py --phase 3 --dataset large --lang fr 

The structure channel uses result of name channel to get name-based seeds. Make sure run name channel first.

To run RREA model:

python main.py --phase 0 --dataset large --lang fr --model rrea --epoch 100 

python main.py --phase 4  --dataset large --lang fr 

Reference to cite when you use LargeEA in a research paper:

@article{largeEA,
  author    = {Congcong Ge and
               Xiaoze Liu and
               Lu Chen and
               Baihua Zheng and
               Yunjun Gao},
  title     = {LargeEA: Aligning Entities for Large-scale Knowledge Graphs},
  journal   = {{PVLDB}},
  volume    = {15},
  number    = {2},
  pages     = {237--245},
  year      = {2022}
}

We use the codes of MRAEA, RREA, GCN-Align, DGMC, AttrGNN, OpenEA, EAKit, SimAlign.

We also provide the modified version of OpenEA in order to run experiments on RTX3090 GPU:OpenEA-TF2