Sentiment Analysis by Deep Learning Techniques

This study focuses on sentiment analysis in Arabic texts using the embedding approach of AraBERT and various neural network architectures, including CNN, LSTM, GRU, BI-LSTM, and BI-GRU. The objective was to predict the sentiments associated with each comment. According to the findings, combining the embedding of AraBERT with the GRU and BI-GRU models demonstrated superior performance in precision, recall, F1 score, and accuracy measures. Nonetheless, it's worth noting that the AraBERT + LSTM model distinguishes itself due to its notably shorter learning time. This research opens new perspectives for sentiment analysis in Arabic using deep learning.