Arabic Aspect Category Detection Using Traditional Neural Networks and Arbert

Sentiment analysis is one of the most active research fields in natural language processing. The increasing use of Arabic on the internet pushes users of socials networks to share their opinions about products and companies. Sentiment analysis is divided into different granularity documents, sentences, and aspects. In this study, we will tackle the aspect-level sentiment analysis for the Arabic language, namely the aspect category detection (ACD) which is a task aiming at identifying and categorizing a sentence or review from a predefined list of categories. Hence, the purpose of this study is to handle the problem of low resources in a dataset by using deep learning (BILSTM,BILSTM-CNN,BILSTM-CRF,BILSTM-BILSTM-CRF), and the model based on transformers like ARBERT. The experimental results show that the model based on transformers outperform the authors based on deep learning, by achieving an F1-score of 71%.