Data augmentation for Arabic text classification: a review of current methods, challenges and prospective directions

Area of Science:

• Natural Language Processing (NLP)
• Machine Learning
• Artificial Intelligence

Background:

• Data augmentation techniques artificially generate new data to improve machine learning model performance.
• These methods address challenges like limited training data and class imbalance in various domains.
• Their application in Arabic text classification is crucial for enhancing classifier performance.

Purpose of the Study:

• To review and analyze data augmentation techniques specifically for Arabic text classification.
• To provide a comprehensive understanding of existing approaches in Arabic Natural Language Processing (ANLP).
• To identify research trends, gaps, and future directions in this specialized field.

Main Methods:

• A systematic review of Arabic studies on data augmentation in text classification published between 2019 and 2024.
• Application of specific inclusion and exclusion criteria to ensure a focused and relevant analysis.
• Synthesis of findings regarding dominant research areas, text types, and dataset availability.

Main Results:

• Research predominantly covers sentiment analysis and propaganda detection, with limited exploration in areas like sarcasm detection.
• A significant lack of benchmark datasets for Arabic text classification tasks was observed.
• Most studies concentrate on short texts (e.g., social media), highlighting a need for research on long texts.

Conclusions:

• Further investigation into data augmentation for Arabic long texts is essential.
• A rigorous comparison of techniques tailored to Arabic's unique linguistic features is required.
• This review offers valuable insights for advancing Arabic NLP and selecting optimal data augmentation strategies.