A robust hybrid approach with product context-aware learning and explainable AI for sentiment analysis in Amazon user reviews

In the ever-changing world of business, gaining valuable insights from customer perspectives is crucial. Consumer evaluations are crucial performance indicators for businesses seeking to enhance their impact. Cyberspace is expanding with an increasing volume of reviews, making it challenging to extract relevant information for desired products. This research explores sentiment analysis for Amazon product reviews in the domain of communication technology, utilizing four publicly available datasets. Sentiment analysis is frequently employed to support E-Commerce platforms in monitoring customer feedback on their products and striving to understand customer needs and preferences. Acknowledging that solely relying on user reviews is insufficient to achieve the best performance, we enhance our approach by incorporating additional context from product titles and headlines for a more comprehensive understanding of the learning algorithm. This paper utilizes three distinct embedding methods, including TF-IDF, Word2Vec, and FastText. FastText outperformed other embeddings when stacked with XGBoost and CatBoost, resulting in the FastXCatStack model. This model achieved accuracy scores of 0.93, 0.93, and 0.94 on mobile electronics, major appliances, and personal care appliances datasets respectively, and linear SVM showed an accuracy score of 0.91 on software reviews when combined with FastText. This research study also provides a comprehensive analysis of deep learning-based models, including approaches like LSTM, GRU, and convolutional neural networks as well as transformer-based models such as BERT, RoBERTa, and XLNET. In the concluding phase, interpretability modeling was applied using Local Interpretable Model-Agnostic Explanations and Latent Dirichlet Allocation to gain deeper insights into the model's decision-making process.