## Usage
Training and local alignment:
--mt_pair: use pseudo pairs
--mt_anchor_weight: use globally guided loss
--weight_decay: the weight decreasing mechanism of globally guided loss
--data: path of data, e.g.: ./data/DBP15K or ./data/MedED
--rate: #training_pairs / (#training_pairs + #testing_pairs)
--cuda: gpu or cpu

example 1:
python UED.py --mt_pair --mt_anchor_weight --weight_decay --lang zh_en --data ./data/DBP15K --rate 0.3 --cuda
example 2:
python UED.py --mt_pair --mt_anchor_weight --weight_decay --lang es_en --data ./data/MedED --rate 0.3



Global alignment method and testing:
python OTP.py --log_path train_logmt_pair_mt_anchor_weight_weight_decay_rate0.3 --lang zh_en --data ./data/DBP15K --rate 0.3


## Datasets:
a. For DBP15K, we add mt_pair99, mt_sim_topK3 to the dataset we inherit from previous works (See data.zip and our paper for details). 

b. For MedED, due to copyright of UMLS, we have hidden the real name of each entity and relationship in MedED, only represented by an id. If you need to obtain the real information of the entity and relationship, or construct KGs of other sizes, please download UMLS (version=2019ab, at https://www.nlm.nih.gov/research/umls/archive/archive_home.html), and refer to our data generation code in data.zip.

Due to the limitaion of size, the data containing DBP15K and MedED can be download at the anonymous link: https://anonymshare.com/OEya/data.zip


#open source#
We will release the source code and datasets after the review。