Revealing and Predicting Online Persuasion Strategy with Elementary Units

This is an official implementation for our paper entitled "Revealing and Predicting Online Persuasion Strategy with Elementary Units" at EMNLP 2019.

ChangeMyView Annotated Dataset

Please download the ChangeMyView annotated dataset (.zip) below:

• http://katfuji.lab.tuat.ac.jp/nlp_datasets/

Unzip and place the downloaded dataset in the top directory. Or, you can setup by:

./setup_dataset.sh

Run

Environment

Requires Python 3.5+, and following packages:

pip install -r requirements.txt

Prepare the dataset and GloVe

Run following script to download the dataset and GloVe embedding file.

./setup_dataset.sh

Run training script

python blc_trainer.py --gpu {GPU No. or -1 if using CPU}

Optionally, you can use GPUs by installing cupy. https://docs-cupy.chainer.org/en/stable/install.html

You will see log directory when training run. The log directory contains parameter log, evaluation results, sample test output, and loss logs. Note that evaluation results change for each run because we use random train/test split and initializations per run.

Citation

If you use our dataset or code, please cite following papers.

1. [EMNLP 2019] Revealing and Predicting Online Persuasion Strategy with Elementary Units

This paper is a pilot study which analyzes EUs and shows the method to predict EUs by neural networks.

@inproceedings{morio-etal-2019-revealing,
    title = "Revealing and Predicting Online Persuasion Strategy with Elementary Units",
    author = "Morio, Gaku  and
      Egawa, Ryo  and
      Fujita, Katsuhide",
    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/D19-1653",
    pages = "6275--6280",
}

2. [ACL-SRW 2019] Annotating and Analyzing Semantic Role of Elementary Units and Relations in Online Persuasive Arguments

This paper describes detailed annotation process of arguments (EUs and Support/Attack links) in ChangeMyView and conducts some examinations.

@inproceedings{egawa-etal-2019,
    title = "Annotating and Analyzing Semantic Role of Elementary Units and Relations in Online Persuasive Arguments",
    author = "Egawa, Ryo  and
      Morio, Gaku  and
      Fujita, Katsuhide",
    booktitle = "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop",
    month = jul,
    year = "2019",
    address = "Florence, Italy",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/P19-2059",
    pages = "422--428",
}