multilingual_irony

URL: https://github.com/bilalghanem/multilingual_irony

Description: Irony detection in a multilingual context.

Dataset

• Name: Arabic Irony Corpus
• Size:
  • ECIR dataset:
    • Training data: 240 KB
    • Testing data: 23.6 KB
  • IDAT dataset:
    • Training data: 94.3 KB
    • Testing data: 23.6 KB

Additional Information

How the datasets were created

This dataset is part of the research presented in the Irony Detection in a Multilingual Context (ECIR-2020) paper.

• The Arabic Irony Corpus consists of approximately 5,500 tweets, manually annotated by two native Arabic speakers.
• An additional randomly sampled 5,500 tweets were appended from the original unannotated corpus for the ECIR dataset (ECIR_training.csv & ECIR_test.csv).
• The dataset was also used in the IDAT shared task at FIRE-2019, but without the additional random samples to maintain data quality (IDAT_training.csv & IDAT_test.csv).

Due to Twitter policy, only tweet IDs are distributed. A Python script (read_tweets_text.py) is provided to retrieve the original tweet texts.

Training methods applied

Information not available.

Results obtained

Information not available.