Introduction

Dataset (ToSarcasm) and Code (TOSPrompt) for CCL 2022 best paper: 面向话题的讽刺识别:新任务、新数据和新方法(Topic-Oriented Sarcasm Detection: New Task, New Dataset and New Method)

Bin Liang (梁斌), Zijie Lin (林子杰), Bing Qin (秦兵), Ruifeng Xu* (徐睿峰)

ToSarcasm

This is a chinese dataset for topic-oriented sarcasm detection. The form of each sample is:

{
  'topic': '美国两党被曝都曾花钱挖特朗普黑料', 
  'text': '真的很优秀',
  'label': '1'
}

where 0-非讽刺(non-sarcasm), 1-讽刺(sarcasm).

⭐ We have relabeled the data, and the statistical information of re-labeled dataset is as follows:

ToSarcasm	Train	Dev	Test
Sarcasm	1608	678	623
Non-Sarcasm	1317	295	350
All	2925	973	973

For more details on the dataset please see the paper.

TOSPrompt

A prompt-based approach for topic-oriented sarcasm detection.

Dependencies

Python 3.7.10
PyTorch 1.11.0
transformers 4.18.0
OpenPrompt 0.1.1

Getting Start

You can easily clone this project via:

git clone https://github.com/HITSZ-HLT/ToSarcasm.git
cd ToSarcasm

Then, create a runnable conda environment via:

conda env create -f environment.yml
source activate && conda activate tosprompt

Data

We evaluate our model on our proposed dataset ToSarcasm.

The data for experiment are placed in ./data/.

Train & Test

python run_prompt.py \
--do_train \
--do_eval \
--epoch 6 \
--batch_size 32 \
--max_length 256 \
--lr 2e-5 \
--weight_decay 2e-3 \
--data_path ./data/ \
--seeds 42 \
--gpu_id 0

if you are running in a non-cuda environment, you can set --gpu_id -1 to run with cpu.

Citation

@inproceedings{liang-etal-2022-mian,
    title = "面向话题的讽刺识别:新任务、新数据和新方法(Topic-Oriented Sarcasm Detection: New Task, New Dataset and New Method)",
    author = "Liang, Bin  and
      Lin, Zijie  and
      Qin, Bing  and
      Xu, Ruifeng",
    booktitle = "Proceedings of the 21st Chinese National Conference on Computational Linguistics",
    year = "2022",
    url = "https://aclanthology.org/2022.ccl-1.50",
    pages = "557--568"
}