📖 DGCN_EL: This is the source code for WWW20' paper: Dynamic Graph Convolutional Networks for Entity Linking
Python 2.7
tensorflow>=1.14
- insert wiki-cross and yago dictionary into mongodb
python entity/mongo_utils.py #you also need to run add_index() to fasten the candidate generation process
- change 'Project Absolute Path' into your project pathes; generate candidates
python s2_1_get_mention_cands.py
python s2_2_get_e_mention_cands.py
- check our generated candidates with EMNLP2017
python s3_1_check_entity_cands.py
- get all entities
python s3_2_get_entity_2_id.py
- extract entity Ganea embeddings, transE embedding, word2vec embedding
python s3_3_get_entity_2_embed.py
- extract entity notable type from freebase
python s3_4_get_entity_type.py
- extract all words embedding
python steps/s4_gen_word_infos.py
- insert training data into the mongodb
python embeddings/get_ent_linking_data.py
python train_global_gen_elmo.py --test=False --data_source=EntityLinking \
--margin_param=0.1 --learning_rate_start=1e-4 \
--loss_type=margin --model_type=SimpleCNNLocalEntLinkModel \
--epoch=150 --lbp_iter_num=2 --keep_prob=0.8 --WS_weight=WS \
--gcn_activation=tanh --gcn_hidden_V=True --l2_w=0.0 --reg_w=0.0 \
--use_unk=none --keep_prob_V=1.0 --residual_w=1.0 --keep_prob_D=0.5 \
--A_diag=dense --start_num=0 --re_train_num=0 --optimizer=normal \
--cand_nums=30 --data_tag_number=0 --data_item_number=0
bash local.sh
bash get_repruned_cand.sh - re-prune candidates
bash get_pretrained_ctx_score.sh - get pre-trained context score
bash global.sh
If you find this project useful, consider citing our paper:
@inproceedings{DBLP:conf/www/WuZMGSH20,
author = {Junshuang Wu and
Richong Zhang and
Yongyi Mao and
Hongyu Guo and
Masoumeh Soflaei and
Jinpeng Huai},
title = {Dynamic Graph Convolutional Networks for Entity Linking},
booktitle = {{WWW} '20: The Web Conference 2020, Taipei, Taiwan, April 20-24, 2020},
pages = {1149--1159},
publisher = {{ACM} / {IW3C2}},
year = {2020},
url = {https://doi.org/10.1145/3366423.3380192},
doi = {10.1145/3366423.3380192}
}