DeepASA: Arabic Sentiment Analysis using Deep Learning and Ensemble Learning

Classifying sentiments held in users’ generated text to favorable, unfavorable (positive/negative) or neutral based on its textual content can offer huge opportunities. In fact, it can aid in many real-life applications like decision making, competitive analysis, strategies planning, identifying trends and rumors, and grasping consumers’ needs. Therefore, the correct analysis and classification of these opinions are decisive, which can have a huge impact on many levels. Many studies exploit machine learning techniques to tackle the sentiment analysis (SA) task. However, after the recent prosperity of deep learning (DL) in many fields, data scientists and researchers have widely utilized it to solve the task of SA. Here, we address the SA task for Arabic text by proposing a model, Deep learning for Arabic Sentiment Analysis (DeepASA), based on two types of recurrent networks, Gated Recurrent Unit (GRU), and Long Short-Term Memory (LSTM), with ensemble learning at the final layer. The performance of DeepASA was evaluated on six different datasets, reviews and tweets, and the final results demonstrated that DeepASA outperformed the start-of-the-art results in all the selected datasets. In fact, the classification accuracy achieved by DeepASA reached 94.32%, and also it was able to reduce the classification error rate by up to 26%