Enhancing Arabic Sentiment Analysis Through a Hybrid Deep Learning Approach
Mustafa Mhamed, Jamal Ali Noja · 2023
Sentiment analysis is a key procedure in many natural language processing systems that extract emotions from textual input. In recent years, Arabic sentiment analysis has become a significant study area. With the growth of social media platforms and data flow, especially in Arabic, substantial difficulties have emerged that call for new strategies to address problems, such as the Arabic language's complicated development and the complexity of the multiple, binary, or massively imbalanced Arabic dataset categorizations. Besides, the system's limitations, whether in online analysis tools, deep or machine learning. This paper proposes a new conjunction method for Arabic sentiment analysis (ASA) called Hybrid Convolution Gate Long (HCGL). This method allows us to extract the best features, handle sequences of different lengths to capture context, address the issue of disappearing error gradients, and improve prediction performance. To match other research works, we conduct studies using a variety of data splits. Furthermore, we pay great attention to Arabic preparation by using all-encompassing procedures that help us address the Arabic language context. The proposed method performs highest in terms of 2-class way efficiency (95.88%), followed by 3-class way performance (95.92%). Additionally, we apply it to the massive Arabic sentiment dataset; it performs well, achieving 88.40% in less time.