Hate speech detection in the Arabic language: corpus design, construction, and evaluation

Area of Science:

• Natural Language Processing
• Machine Learning
• Computational Linguistics

Background:

• Arabic hate speech detection is complex due to linguistic diversity.
• Existing methods lack comprehensive Arabic hate speech datasets.
• Online platforms require effective tools to combat harmful content.

Purpose of the Study:

• To introduce a novel, large-scale, multi-class Arabic hate speech dataset.
• To evaluate machine learning model performance for Arabic hate speech identification.
• To facilitate further research in Arabic online content moderation.

Main Methods:

• Developed a public dataset of 403,688 annotated Arabic tweets.
• Utilized text representation models: Word2Vec, TF-IDF, and AraBert.
• Evaluated seven machine learning classifiers: SVM, LR, NB, RF, AdaBoost, XGBoost, and CatBoost.

Main Results:

• The novel dataset proved efficient for hate speech detection tasks.
• Encouraging assessment outcomes were achieved in challenging, unstructured text.
• The dataset supports robust evaluation of various machine learning models.

Conclusions:

• The created dataset is a valuable resource for Arabic hate speech research.
• Machine learning models demonstrate potential for effective Arabic hate speech identification.
• Further academic investigation into this critical area is now enabled.