<p>Product review sentiment analysis is challenging due to noisy user-generated content, domain-specific language, multimodal signals and platform-specific sentiment expression styles. This paper proposes EMOTEX-PR, a variant of the EMOTEX framework for strong, real-time sentiment analysis of product reviews over various datasets. The approach improves every step in the sentiment analysis pipeline. In the pre-processing stage, the Emotion-Preserving Contextual Pre-processing (EP-CP) technique carries out domain-aware normalization while preserving sentiment conveying words and context-dependent product-specific aspects without over-simplifying reviews. In feature extraction, the Multimodal Emotion-Aware Feature Extraction (MEAFE) module combines textual embedding’s with auxiliary signals like acoustic tone (for verbal reviews) and visual sentiment signals (from video-based feedback) to allow a more detailed representation of user sentiment. The classification phase uses an Optimized Transformer-Augmented Dynamic Attention Network (OT-EDA) to dynamically attend to important sentiment indicators and adjust to dataset-specific features. The optimization phase, Multi-Objective Real-Time Optimization (MOR-TO) applies Genetic Algorithms (GA) for architecture selection and Reinforcement Learning (RL) for fine-tuning and it balances accuracy and latency in a trade-off. Experimental evaluation on benchmark product review datasets such as Amazon, Yelp, IMDB, Flipkart and these combined dataset shows that EMOTEX-PR outperforms other state-of-the-art models in terms of accuracy (98.3%), F1-score of 98.1, latency of 95ms and fairness index of 0.88. The framework is suitable for deployment in large-scale e-commerce analytics with application to real-time customer feedback monitoring and decision support.</p>