isakzhang / absa-qa

ABSA-QA

Source code for our paper "Aspect-based Sentiment Analysis in Question Answering Forums" in EMNLP-Findings 2021.

Requirements

• Python 3.7.4
• torch 1.6.0
• fastNLP 0.5.5
• nltk 3.5

Data

The datasets are from: Aspect Sentiment Classification Towards Question-Answering with Reinforced Bidirectional Attention Network (ACL 2019)

*Due to the unclear license of the original data, we do not include it here, pls download from the original link in the paper.

Training and Evaluation

To train the model:

python train_model.py --dataset $category

where $category in [elec, beauty, bags] denotes the dataset name, the default one is the elec dataset.

The model with best performance on the dev dataset will be saved in outputs/ and evaluated on the test data after the training phase is finished

Citation

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

@inproceedings{zhang-etal-2021-aspect,
    title = "Aspect-based Sentiment Analysis in Question Answering Forums",
    author = "Zhang, Wenxuan  and
      Deng, Yang  and
      Li, Xin  and
      Bing, Lidong  and
      Lam, Wai",
    booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2021",
    month = nov,
    year = "2021",
    address = "Punta Cana, Dominican Republic",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2021.findings-emnlp.390",
    pages = "4582--4591"
}

absa-qa's Issues

Model class does not contain ``asc_target_vocab`` param

In the train_model.py module the Model is called with the asc_target_vocab parameter which does not actually feature in the Model class in model.py

Access to dataset

Congratulations on your great work about the ABSA task of QA forum, the original link of the dataset has been 404 not found, could you offer a valid link or dataset to me? Thanks a lot.