Integrating Sentiment Analysis and Topic Modeling for Actionable Business Intelligence: A Comparative Study of Machine Learning and Deep Learning Architectures for Arabic Company Reviews

Abstract

This study presents a comprehensive framework for Arabic Sentiment Analysis (ASA) integrated with Topic Modeling to support actionable Business Intelligence (BI) from company reviews. We conduct a comparative evaluation of traditional Machine Learning (ML) classifiers (Logistic Regression, Support Vector Machine, Naive Bayes, and Random Forest) and Deep Learning (DL) architectures (Bidirectional LSTM and Bidirectional GRU). Experiments reveal that no single model dominates across all evaluation metrics. Logistic Regression achieved the highest Macro F1-score (0.6257), indicating the most balanced performance under class imbalance. Naive Bayes obtained the highest accuracy (0.8385) and weighted F1-score (0.8175), reflecting strong performance on the dominant sentiment class. Deep learning models achieved competitive accuracy but did not surpass linear models in Macro F1. Topic Modeling using Latent Dirichlet Allocation (LDA) complements sentiment classification by extracting interpretable strengths and weaknesses for each company, transforming sentiment analysis into a practical BI tool.