Skip to content

Latest commit

 

History

History
80 lines (49 loc) · 2.3 KB

README.md

File metadata and controls

80 lines (49 loc) · 2.3 KB

EASY

Source code for Make it Easy: An Effective End-to-End Entity Alignment Framework. SIGIR 2021.


Overview

Installation

To run our code, first install required packages. Then run preprocess

pip install -r requirements.txt
sh preprocess.sh

Run

Run on all dataset with default settings

First get NEAP results.

sh neap.sh

Then get SRS results.

python main.py --pair all

Run on specific dataset/settings

NEAP

The SRS process need the result of NEAP. To get NEAP results on a specific dataset(e.g. en_fr)

python neap.py --pair en_fr

For fasttext, please download aligned word vectors wiki.{lang}.align.vec and place them into aligned_vectors/ folder.

mkdir aligned_vectors
cd aligned_vectors
wget https://dl.fbaipublicfiles.com/fasttext/vectors-aligned/wiki.en.align.vec
wget https://dl.fbaipublicfiles.com/fasttext/vectors-aligned/wiki.fr.align.vec
wget https://dl.fbaipublicfiles.com/fasttext/vectors-aligned/wiki.de.align.vec

SRS

After acquiring similarity matrices from NEAP, run main.py to refine.

python main.py --pair en_fr

Change arguments for different settings. To get help on arugments, run

python main.py --help

The refinement process is based on the code of MRAEA, RREA, GCN-Align. In our experiment, training is done on CPU.

Acknowledgement

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

DBP15k dataset is from GMNN and AttrGNN.

SRPRS dataset is from RSN.