Exploring the Use of Word Embedding and Deep Learning in Arabic Sentiment Analysis

In the past couple of years, improving Arabic sentiment Analysis systems has been one of the important fields of research. There are several challenges and issues facing existing systems, especially, handling multiple dialects and feature extraction. Most of those systems are generated using linear classification models and traditional bag-of-word features. In this work, we explore the use of word embedding as a modern feature representation, and Convolutional Neural Networks as a Deep Neural Network in a sentiment classification of Arabic texts. The application of our model on five benchmark datasets has yielded results that outperform previous works on 4 out of 5 datasets. © 2020, Springer Nature Switzerland AG.