Natural Language Processing and Sentiment Ana...

Natural Language Processing and Sentiment Analysis for Libyan Arabic Language Dataset

Author Name : Mohamed Muftah Abubaera, Salman Mohammed Jiddah

DOI: https://doi.org/10.56025/IJARESM.2024.126242753

ABSTRACT

This study presents a novel approach for sentiment analysis on the Libyana telecommunication user sentiment dataset, leveraging Natural Language Processing (NLP) and a hybrid Convolutional Neural Network–Gated Recurrent Unit (CNN-GRU) model. Sentiment analysis is crucial in understanding public opinions on various subjects, and this research specifically addresses the complexities of Arabic sentiment analysis, characterized by diverse dialects and slangs. The proposed approach integrates data augmentation through random text insertion to increase dataset size and balance, thereby enabling more effective training of the deep learning model. Two experiments were conducted to evaluate the impact of data augmentation on model performance. The first experiment, using preprocessed data without augmentation, achieved an accuracy of 82.19%. In contrast, the second experiment, incorporating data augmentation, significantly improved accuracy to 91.54%. This nearly 10% improvement underscores the effectiveness of data augmentation in enhancing sample size and addressing the issue of under sampled positive data samples, which is critical for robust model training. The results demonstrate substantial performance gains over previous methods, highlighting the potential of data augmentation in advancing sentiment analysis for Arabic text data.

Keywords: Arabic Text Processing, CNN-GRU Model, Data Augmentation, Natural Language Processing (NLP), Sentiment Analysis.