Overview

Here is the source code for our NAACL 2022 paper A Dual-Channel Framework for Sarcasm Recognition by Detecting Sentiment Conflict.

Run the scripts

Please modify the parameters in train.sh and run sh train.sh.

Experiments

* Datasets

We evaluate our model on three benchmark datasets, IAC-V1, IAC-V2, Twitter. The split datasets are in /data.

The official websites for datasets:

• IAC-V1: https://nlds.soe.ucsc.edu/sarcasm1
• IAC-V2: https://nlds.soe.ucsc.edu/sarcasm2
• Twitter: https://github.com/Cyvhee/SemEval2018-Task3

* Main results

Model	IAC-V1		IAC-V2		Tweets
	F1	Acc.	F1	Acc.	F1	Acc.
UCDCC	58.5	58.5	67.0	67.0	72.4	79.7
THU-NGN	64.2	64.3	73.3	73.3	70.5	73.5
Bi-LSTM	64.6	64.6	79.7	79.7	71.7	73.0
At-LSTM	65.3	65.5	76.1	76.2	70.0	70.2
CNN-LSTM-DNN	60.9	61.1	75.2	75.3	71.9	72.3
MIARN	64.9	65.2	75.2	75.3	68.8	70.2
ADGCN	64.3	64.3	80.9	80.9	72.8	73.6
DC-Net (Ours)	66.4	66.5	82.1	82.1	76.3	76.7

Citation

@inproceedings{DBLP:conf/naacl/LiuWSMLG22,
  author    = {Yiyi Liu and
               Yequan Wang and
               Aixin Sun and
               Xuying Meng and
               Jing Li and
               Jiafeng Guo},
  title     = {A Dual-Channel Framework for Sarcasm Recognition by Detecting Sentiment
               Conflict},
  booktitle = {Findings of the Association for Computational Linguistics: {NAACL}
               2022, Seattle, WA, United States, July 10-15, 2022},
  pages     = {1670--1680},
  publisher = {Association for Computational Linguistics},
  year      = {2022},
  url       = {https://doi.org/10.18653/v1/2022.findings-naacl.126},
  doi       = {10.18653/v1/2022.findings-naacl.126},
  biburl    = {https://dblp.org/rec/conf/naacl/LiuWSMLG22.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}
}