Heterogeneous Ensemble Deep Learning Model for Enhanced Arabic Sentiment Analysis.

Area of Science:

• Natural Language Processing
• Artificial Intelligence
• Computational Linguistics

Background:

• Sentiment analysis is crucial for understanding opinions on social media.
• Arabic sentiment analysis faces challenges due to linguistic complexity and dialectal variations.
• Existing machine learning and deep learning models show limitations in accuracy for Arabic text.

Purpose of the Study:

• To propose an optimized heterogeneous stacking ensemble model for enhanced Arabic sentiment analysis.
• To improve the prediction accuracy of sentiment analysis for Arabic content.
• To address the limitations of current sentiment analysis mechanisms for the Arabic language.

Main Methods:

• Developed a heterogeneous stacking ensemble model integrating pre-trained Deep Learning (DL) models (RNN, LSTM, GRU) with meta-learners (LR, RF, SVM).
• Optimized DL model parameters using KerasTuner and Machine Learning (ML) parameters via Grid search.
• Compared the proposed ensemble model against individual DL models and traditional ML techniques (DT, LR, KNN, RF, NB) on three Arabic datasets.

Main Results:

• The proposed optimized heterogeneous stacking ensemble model achieved superior performance across all evaluated metrics (accuracy, precision, recall, f1-score).
• The ensemble model demonstrated significant improvements compared to individual DL models and standard ML algorithms.
• Consistent best performance was observed for the ensemble model on all benchmark Arabic datasets.

Conclusions:

• The optimized heterogeneous stacking ensemble model effectively enhances Arabic sentiment analysis performance.
• This approach offers a robust solution for overcoming the complexities of Arabic sentiment analysis.
• The findings suggest a promising direction for future research in cross-lingual and dialectal sentiment analysis.