A BiGRU-Based Model Augmented with Attention for Arabic Aspect-Based Sentiment Analysis

Individuals in today’s digital world have the ability to share their opinions widely owing to the internet and social media platforms. Aspect-Based Sentiment Analysis (ABSA) is an effective technique helps in comprehending these opinions. Accordingly, this study proposes an Arabic aspect sentiment classification model that incorporates the state-of-the-art natural language processing (NLP) techniques. The model was constructed using three layers: an embedding layer that used an Arabic transformer-based pre-trained language model. Subsequently, bidirectional gated recurrent unit (Bi-GRU) layer was added for capturing contextual information in the given sentence. On top of the Bi-GRU, a soft attention layer was used to draw the model’s attention to the important words. The benchmark Arabic hotel reviews dataset was used for evaluating the proposed model. The evaluation finding demonstrate that the proposed model showed a 14.41% enhancement in accuracy over the baseline model, and exceeded state-of-the-art models by 1.3%, attaining an accuracy of 90.81%.