Saudi-Dialect-Irony-Dataset

URL: https://github.com/iwan-rg/Saudi-Dialect-Irony-Dataset

Description: The Saudi irony dataset was collected using the Twitter API and consists of 19,810 tweets, 8,089 of them are labeled as ironic tweets.

Dataset

• Name: Saudi Dialect Irony Dataset (Sa`7r ساخر)
• Size: 19,810 tweets
  • 8,089 ironic tweets
• File:
  • SaudiIrony.csv: Contains tweets with two labels: “ironic” and “non-ironic” (3.1 MB).

Additional Information

How the datasets were created

The Saudi irony dataset (Sa`7r ساخر) was collected using the Twitter API. It consists of 19,810 tweets, out of which 8,089 are labeled as ironic. This dataset was created as part of a study on irony detection in sentiment analysis.

Training methods applied

The dataset was used to train several machine learning models for irony detection:

• Machine learning models:
  • K-Nearest Neighbor (KNN)
  • Logistic Regression (LR)
  • Support Vector Machine (SVM)
  • Naïve Bayes (NB)
• Deep learning models:
  • BiLSTM
  • AraBERT

Results obtained

• SVM achieved the highest accuracy among the machine learning models with an accuracy of 0.68.
• BiLSTM achieved an accuracy of 0.66.
• AraBERT achieved the highest accuracy overall with 0.71 for detecting irony phrases in Saudi Dialect.

Read the full paper.