Tianbaojie / ACS

Cross-lingual ABSA (XABSA)

This repo contains the data and code for our paper Cross-lingual Aspect-based Sentiment Analysis with Aspect Term Code-Switching in EMNLP 2021.

Requirements

• torch==1.3.1
• numpy==1.19.4
• transformers==3.4.0 (You can also use the static version in this repo)
• sentencepiece==0.1.91
• tokenizer==0.9.2
• sacremoses==0.0.43

Quick Start

• Download the pre-trained multilingual language model mBERT or XLM-R
• To quickly reproduce the results with French (fr) as the target langauge and mBERT (mbert) as the backbone under the supervised setting:

python main.py --tfm_type mbert --tgt_lang fr

• To reproduce other results, ref to the data_utils.py for details

Usage

To run experiments under different settings, change the exp_type setting:

• supervised refers to the supervised setting
• acs is the proposed method
• acs_kd_s/m: single/multi teacher distillation under the cross-lingual setting
• macs_kd: multilingual distillation

Two example scripts:

• run_absa.sh provides an example to run basic experiment.
• run_absa_kd.sh provides an example to run experiments with knowledge distilling on the unlabeled target laguage data.

Citation

If the code is used in your research, please star our repo and cite our paper as follows:

@inproceedings{zhang-etal-2021-cross,
    title = "Cross-lingual Aspect-based Sentiment Analysis with Aspect Term Code-Switching",
    author = "Zhang, Wenxuan  and
      He, Ruidan  and
      Peng, Haiyun  and
      Bing, Lidong  and
      Lam, Wai",
    booktitle = "Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2021",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.emnlp-main.727",
    pages = "9220--9230",
}

​