Machine Learning Tools In Sentiment Analysis of Arabic Social Media

Abstract

This thesis focuses on sentiment analysis on Arabic social media, which is a challenging task due to the complex and nuanced nature of Arabic language. To address this challenge, the thesis proposes several contributions, including the development of a new Arabic word embedding model, enhancements to the BERT model for imbalanced text classification. The proposed Arabic word embedding model is designed to capture the unique semantic relationships and nuances of Arabic language, which can improve the accuracy of sentiment analysis. The enhancements to the BERT model include modifications to the attention mechanism and training process, which can improve the performance of the model on Arabic text data. The inclusion of a balancing layer in BERT approaches is aimed at addressing the issue of imbalanced data, which is common in sentiment analysis tasks. The thesis presents experimental results that demonstrate the effectiveness of the proposed contributions in improving the accuracy and performance of sentiment analysis on Arabic social media. The proposed model outperforms the BERT baseline and achieves state-of-the-art results on several benchmark datasets.