Skip to content

h324yang/HMAN

Folders and files

NameName
Last commit message
Last commit date

Latest commit

44f1fb4 · Nov 25, 2019

History

17 Commits
Sep 25, 2019
Nov 14, 2019
Oct 20, 2019
Oct 20, 2019
Sep 25, 2019
Sep 25, 2019
Nov 25, 2019
Oct 20, 2019
Oct 20, 2019
Oct 20, 2019
Oct 15, 2019
Oct 20, 2019
Oct 20, 2019

Repository files navigation

Hybrid Multi-Aspect Alignment Networks (HMAN)

The code for the research paper Aligning Cross-lingual Entities with Multi-Aspect Information @EMNLP 2019.

*Our code is built on top of GCN-Align.

Quick Demo

We stored the embeddings for demo in the directory demo_embd, and we can evaluate ZH-EN as follows:

python weighted_concat.py -d demo_embd/pairwise_dump.json -g demo_embd/zh_en_graph_embd.pkl -i data/zh_en/test

Run Graph-based Embeddings (HMAN/MAN)

HMAN

bash graph.sh 0 1

MAN

bash graph.sh 0 0

Run PairwiseBERT

We use the package relogic to derive BERT-based embeddings. Please install relogic first:

git submodule init
git submodule update

Also, becasuerelogic evolves fast, we suggest that you change to the commit d1b5046:

cd relogic
git checkout d1b5046

Note that the argument --local_rank in relogic indicates your gpu id.

Training

bash train_bert.sh 0 zh_en

(Note that you need to stop the training manually.)

Evaluation

bash eval_bert.sh 0 zh_en

Integration

python weighted_concat.py --desc relogic/saves/pair_matching/zh_en/pairwise_dump.json --graph graph_ckpt/zh_en_graph_embd.pkl --ill data/zh_en/test

Citation

@article{yang2019aligning,
  title={Aligning Cross-Lingual Entities with Multi-Aspect Information},
  author={Yang, Hsiu-Wei and Zou, Yanyan and Shi, Peng and Lu, Wei and Lin, Jimmy and Sun, Xu},
  booktitle={Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing},
  year={2019}
}

About

Hybrid Multi-Aspect Alignment Networks

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published