Exploring Arabic Hotel Reviews: Sentiment Insights Through Deep Learning and BERT Transformers Models

An application of natural language processing (NLP) called sentiment analysis (SA) has been a more popular field of study in recent years in both scientific research and the business world. Finding the opinion stated in a subjective statement with numerous methods demonstrating some degree of success in identifying sentiment across numerous datasets is the goal of this activity. An opinion is a personal statement of one’s subjective feelings and beliefs. These sensations and ideas might be associated with a certain emotion. Two sentiments have been examined the most: negative and positive. Deep learning and transformer models are the two most recent advancements in sentiment analysis techniques since the attention mechanism in machine learning was introduced. In recent times, transformer-based models, which are attentive encoder-decoder systems, have demonstrated remarkable performance in sentiment analysis and have been regarded as the most advanced model for a variety of languages. Utilizing this property in encoder-decoder architecture has affected transformer models’ efficacy in sentiment analysis and other NLP tasks. Even so, there is still room for further enhancement of Arabic sentiment analysis efficiency. As a result, bidirectional encoder representations from transformers (BERT) and deep learning (DL) models such as convolutional neural networks (CNN) and bidirectional long-short-term memory (bi-LSTM) have been tested in this work. To achieve better performance, an ensemble model has also been incorporated. By including the standard BERT tokenizer and transformer models, the suggested method is improved. Different models were tested. When compared to CNN and Bi-Lstm, the experimental investigation demonstrates the effectiveness of the suggested strategy in terms of realistic performance.