ArSarcasm

URL: https://github.com/iabufarha/ArSarcasm

Description: This repository contains the Arabic sarcasm dataset (ArSarcasm).

Project Overview

ArSarcasm is a dataset for sarcasm detection in Arabic tweets. It was built using existing Arabic sentiment analysis datasets (SemEval 2017 and ASTD) and includes annotations for sarcasm and dialect.

Dataset:

The dataset contains 10,547 tweets, where 1,682 (16%) are labeled as sarcastic. The dataset is available in CSV format with an 80/20 train-test split:

• Training set: 8,437 tweets
• Test set: 2,110 tweets

Dataset Fields:

• tweet: The original tweet text.
• sarcasm: Boolean indicating whether the tweet is sarcastic.
• sentiment: New annotation for sentiment (positive, negative, neutral).
• original_sentiment: Sentiment from the original dataset annotation.
• source: The original dataset (SemEval or ASTD).
• dialect: The Arabic dialect used in the tweet, categorized into:
  • msa: Modern Standard Arabic
  • egypt: Egyptian and Sudanese Arabic
  • levant: Levantine Arabic (Palestine, Jordan, Syria, Lebanon)
  • gulf: Gulf Arabic (Saudi Arabia, UAE, Qatar, etc.)
  • magreb: North African Arabic (Algeria, Libya, Tunisia, Morocco)

Dataset Usage:

The dataset is structured for sarcasm detection research and includes sentiment and dialectal variations, making it useful for broader NLP tasks in Arabic.

Dataset Statistics:

• Total tweets: 10,547
• Sarcastic tweets: 1,682 (16%)
• Train set size: 8,437
• Test set size: 2,110

Training Methods:

No specific training methods are provided in the repository.

Results:

No specific results or performance metrics are provided in the repository.

Dataset Files:

• ArSarcasm_train.csv (1.7 MB)
• ArSarcasm_test.csv (435 KB)