XL2248 / AGDT

Code for "A Novel Aspect-Guided Deep Transition Model for Aspect Based Sentiment Analysis." on EMNLP 2019.

Code and data for A Novel Aspect-Guided Deep Transition Model for Aspect Based Sentiment Analysis

Introduction

The implementation is based on THUMT. Download Glove file and change the path in 'AGDT/thumt/thumt/bin/trainer.py' correspondingly. The dataset we used is from GCAE.

Usage

Training with the following scripts:

• ACSA

bash run_train_14.sh
bash run_train_large.sh

• ATSA

bash run_train_r.sh
bash run_train_l.sh

The result can be found in the path like '/14_agdt-result-0/eval/record'.

Requirements

• tensorflow 1.8.0
• python 2.7

Citation

If you find this project helps, please cite our paper :)

@inproceedings{liang-etal-2019-novel-aspect,
    title = "A Novel Aspect-Guided Deep Transition Model for Aspect Based Sentiment Analysis",
    author = "Liang, Yunlong  and
      Meng, Fandong  and
      Zhang, Jinchao  and
      Xu, Jinan  and
      Chen, Yufeng  and
      Zhou, Jie",
    booktitle = "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)",
    month = nov,
    year = "2019",
    address = "Hong Kong, China",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/D19-1559",
    doi = "10.18653/v1/D19-1559",
    pages = "5568--5579",
}