thunlp / ieajke

Code and data for our paper "Iterative Entity Alignment via Joint Knowledge Embeddings"

IEAJKE

Codes and data for our paper "Iterative Entity Alignment via Joint Knowledge Embeddings"

The code for ITransE and DFB-1 are available now!

Dataset

For DFB-1, we provide four files

• common_entities.txt: alignment seeds;
• common_entities2id.txt: id version of alignment seeds;
• newentity2id.txt: the correspondence of entities and their ids;
• relation2id.txt: the correspondence of relations and their ids;
• triple2id.txt: all triples in FB15K. Note that this is not the original form, but a processed one where entity seeds are already combined as one.

How to run

First, compile src/transE.cpp by: g++ transE.cpp -o transE -pthread -O3 -std=c++11 -march=native

The compilation arguments could be different from platform to platform. Such compilation command is the same as the one in Fast-TransX. If you come across problems, you can refer to the issues of Fast-TransX.

Run ./transE.

To test the result, run test.ipynb with Jupyter Notebook. We also provide pretrained embeddings in entity2vec.bern. You should get "0.8309717616319968 0.6708873480052256 80.05868756908853 9951" finally in the sixth block.

Contact

If you have any problems with my paper and datasets, please send email to [email protected].

ieajke's Issues

请问ptransE的代码那边有？

关于T1、T2，以及开发集的获取

每个数据集只有一个存储triple文件，请问如何获取文中所提到的T1、T2，以及大小为1000开发集？

About the relation embeddings in different kgs

Are the relation embeddings shared across the two KGs in the procedure described in Section 4.2 of the paper?

About DFB-2 and DFB-3

I'm trying to reproduce this model. It's by far just DFB-1 is available mightly. How could I get DFB-2 and DFB-3

关于common_entities

该文件是不是已经包括了kg1和kg2的所有对齐实体对，是不是可以认为该文件以外的任意实体对不存在对齐关系

提示错误：段错误 (核心已转储)

（1）运行命令./transE，提示段错误 (核心已转储)。查了一下是内存问题，双系统用ubuntu测试你的代码，请问怎么解决，谢谢！

（2）运行Fast-TransX项目，运行指令报同样错误：./transE -size 50 -sizeR 50 -input ./data -output ./ -load LOAD ./data -load-binary 0 -out-binary 0 -thread 3 -epochs 5 -nbatches 10 -alpha ALPHA 0.01 -margin MARGIN 1 -note ./