ArSa-Tweets: A novel Arabic sarcasm detection system based on deep learning model.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Artificial Intelligence

Background:

• Sarcasm detection in sentiment analysis (SA) is crucial as sarcasm deviates from literal meaning.
• Arabic SA faces challenges including implicit idioms and a lack of dedicated sarcasm corpora.
• Existing methods struggle with the nuances of Arabic sarcastic expressions.

Purpose of the Study:

• To propose and develop a novel model, ArSa-Tweet, for detecting sarcasm in Arabic tweets.
• To address the limitations of existing Arabic SA approaches by incorporating robust preprocessing and advanced deep learning.
• To create a valuable Arabic sarcasm corpus, ArSa-data, for research and development.

Main Methods:

• Implementation and adaptation of various deep learning (DL) models: LSTM, Multi-headed CNN-LSTM-GRU, BERT, AraBert-V01, and AraBert-V02.
• Application of rigorous preprocessing techniques to enhance data quality before DL model input.
• Development of ArSa-data, a specialized corpus of Arabic tweets for sarcasm analysis.

Main Results:

• The ArSa-Tweet model demonstrated significant improvements in sarcasm detection accuracy.
• Comparative analysis confirmed the superior performance of the AraBert-V02 model within the ArSa-Tweet framework.
• The proposed method achieved the highest accuracy rates across all evaluated metrics.

Conclusions:

• The ArSa-Tweet model, particularly with the AraBert-V02 integration, offers a highly effective solution for Arabic sarcasm detection.
• The ArSa-data corpus provides a valuable resource for advancing research in Arabic sentiment analysis.
• This work contributes to overcoming key challenges in understanding nuanced language in Arabic social media.