Sentiment Analysis in Arabic Social Media: Optimized Performance through CNN-LSTM and BERT

Abstract

Within the domain of text mining and natural language processing, sentiment analysis is currently very relevant: it allows us to understand the opinions or feelings of folks as communicated in texts. Sentiment analysis, which is often directed at English and under-researched for Arabic (due to language complexity, large dialects, and low-quality annotated datasets), Most prior works. To tackle these problems, we propose the hybrid deep learning approach with Convolutional Neural Networks and Long Short-Term Memory (CNN-LSTM) from Scenes to Encoder-Decoder Refinement using Bidirectional Encoder representations from Transformers (BERT) in this research. CNN-LSTM, for example, reaches the state-of-the-art accuracy of 95.5% on our dataset of 330k Arabic product reviews while BERT hits 92%. This research adds to it (1) the development of Arabic deep learning models for sentiment classification accuracy in social media, (2) proof of concept that hybrid deep learning is effective on Arabic text, and (3) helps alleviate the dearth of annotated datasets for this task on an enormous scale using extensive data. This research indicates that deep learning models can surmount several hurdles of Arabic sentiment analysis and serve as a stepping stone for more robust, yet scalable solutions.
Corresponding Author