This research presents a thorough and systematic approach to classifying the sentiment of Amazon product evaluations using a deep learning framework based on a bidirectional reinforcement learning network. Utilising a large-scale dataset that has been structured-preprocessed into sentiment labels, the study effectively incorporates numerical ratings and review language. Semantic embedding using previously trained GloVe vectors to a text normalisation, and meticulous data cleaning are some of the methods used in the study to guarantee that the model can grasp all the nuances and context-dependent information in language. Because it efficiently regulates the bidirectional flow of information, BiLSTM is necessary for the model to learn via past and future word correlations inside a review. To prevent overfitting and enhance generalisation, the design is fine-tuned with the use of nationwide pooling, batch normalization, including dropout layers. Stratified sampling is a must for data division in order to ensure adequate representation across sentiment classes, especially considering the inherent bias in players material. Performance evaluation using metrics like F1-score, recall, accuracy, and precision shows that the suggested model is effective. The accuracy indicator (0.9124) and the F1-score (0.9123) stand out among the others, both of which have very high values. The methodology has demonstrated its value by reliably labelling assessments as positive, neutral, or negative. Full exploratory data analysis also shows that class balance and preprocessing have a major effect on the model\'s performance. Finally, this study lays out a scalable and extensible deep learning-based method that works well in sentiment analysis for e-commerce and customer feedback systems, among other real-world uses.
Introduction
Sentiment analysis, or opinion mining, is a key application of natural language processing (NLP) widely used to analyze customer reviews, especially on platforms like Amazon. It helps businesses gauge customer satisfaction, identify issues, and adjust marketing strategies by classifying reviews as positive, negative, or neutral. Traditional methods relied on manual feature selection and classifiers like Naïve Bayes and SVM, but recent advances in deep learning—such as attention mechanisms, RNNs, and LSTMs—have improved understanding of context and emotional nuance.
Modern NLP models, including BERT and other transformer-based architectures, leverage deep contextual embeddings to better capture sarcasm, negation, and complex semantics. Aspect-based sentiment analysis further refines this by separately evaluating sentiments related to different product features within the same review, providing actionable insights. Challenges such as imbalanced datasets, fake reviews, and mixed languages often require hybrid approaches combining rule-based and AI techniques.
Sentiment analysis is extensively used in business for improving marketing, customer service, and product development by processing large volumes of reviews quickly and accurately. Ethical concerns focus on privacy, bias, and transparency of algorithms.
The literature review highlights recent studies employing a range of models from Word2Vec with SVM to transformer-based approaches like BERT and T5, often showing superior accuracy with deep learning models. Preprocessing steps like tokenization, stemming, and vectorization remain crucial. Research also explores sarcasm detection, dimensionality reduction, and hybrid methods to enhance robustness.
This study proposes building a Bidirectional LSTM-based classifier for Amazon reviews, using extensive data cleaning, embedding with GloVe vectors, and techniques like dropout and batch normalization to improve generalization. Performance will be measured using accuracy, precision, recall, and F1-score, aiming to provide valuable sentiment insights for business decision-making.
Conclusion
Finally, the results demonstrate that a deep learning-based Bidirectional short- and long-term memory (BiLSTM) model can reliably gauge customers\' sentiments regarding Amazon purchases. The study lays a good groundwork for model training by carefully working with a large dataset. This involves cleaning up the text, adding sentiment labels, and using GloVe word embedding. The model can figure out complex contextual dependencies in review text since it has spatial dropout, BiLSTM layers, global max pooling, and dense layers with batch normalization and dropout. The BiLSTM model did quite well on the tests, getting 91.24% of the answers right and doing well on the other criteria as well. This means that it can arrange things quite well and is pretty reliable. The BiLSTM proved far better at handling the sequential and complex character of natural language than older models like Random Forest and CNN. This shows how powerful recurrent architectures are, especially when it comes to identifying emotions across several classes. The results not only show that the model can handle real-world, noisy text data, but they also imply that it could help businesses get useful information from customer evaluations. Overall, this work introduces a sentiment analysis pipeline that is both scalable and accurate and can be utilized on any e-commerce site. This will lead to a better customer experience and product development plans based on data.
References
[1] M. Belal, J. She, and S. Wong, “Leveraging ChatGPT As Text Annotation Tool For Sentiment Analysis,” 2023, [Online]. Available: http://arxiv.org/abs/2306.17177
[2] H. Computational, M. Methods, M. Volume, and A. Id, “Retracted: Development of NLP-Integrated Intelligent Web System for E-Mental Health,” Comput. Math. Methods Med., vol. 2023, pp. 1–1, 2023, doi: 10.1155/2023/9780851.
[3] C. A. License, “Universities?: A Research Based on Artificial Intelligence,” vol. 2022, 2023.
[4] “Amazon Product Reviews.” https://aws.amazon.com/blogs/machine-learning/detect-sentiment-from-customer-reviews-using-amazon-comprehend/ (accessed Jun. 28, 2025).
[5] D. C. Feng, W. J. Wang, S. Mangalathu, and Z. Sun, “Condition Assessment of Highway Bridges Using Textual Data and Natural Language Processing- (NLP-) Based Machine Learning Models,” Struct. Control Heal. Monit., vol. 2023, 2023, doi: 10.1155/2023/9761154.
[6] “Retracted: Sentiment Analysis of Statements on Social Media and Electronic Media Using Machine and Deep Learning Classifiers,” Comput. Intell. Neurosci., vol. 2023, pp. 1–1, 2023, doi: 10.1155/2023/9846879.
[7] D. Mahto, S. C. Yadav, and G. S. Lalotra, “Sentiment Prediction of Textual Data Using Hybrid ConvBidirectional-LSTM Model,” Mob. Inf. Syst., vol. 2022, 2022, doi: 10.1155/2022/1068554.
[8] C. Jin, Z. Song, J. Xu, and H. Gao, “Attention-Based Bi-DLSTM for Sentiment Analysis of Beijing Opera Lyrics,” Wirel. Commun. Mob. Comput., vol. 2022, 2022, doi: 10.1155/2022/1167462
[9] Y. Wen, W. Li, Q. Zeng, H. Duan, F. Zhang, and S. Kang, “Syntactic Knowledge Embedding Network for Aspect-Based Sentiment Classification,” Mob. Inf. Syst., vol. 2022, 2022, doi: 10.1155/2022/1352028.
[10] V. Dogra et al., “A Complete Process of Text Classification System Using State-of-the-Art NLP Models,” Comput. Intell. Neurosci., vol. 2022, 2022, doi: 10.1155/2022/1883698.
[11] A. Vitetta, “Sentiment Analysis Models with Bayesian Approach: A Bike Preference Application in Metropolitan Cities,” J. Adv. Transp., vol. 2022, 2022, doi: 10.1155/2022/2499282.
[12] H. Wang, “Word2Vec and SVM Fusion for Advanced Sentiment Analysis on Amazon Reviews,” Highlights in Science, Engineering and Technology, vol. 85. pp. 743–749, 2024. doi: 10.54097/sw4pft19.
[13] M. K. Shaik Vadla, M. A. Suresh, and V. K. Viswanathan, “Enhancing Product Design through AI-Driven Sentiment Analysis of Amazon Reviews Using BERT,” Algorithms, vol. 17, no. 2, 2024, doi: 10.3390/a17020059.
[14] Y. M. Latha and B. S. Rao, “Amazon product recommendation system based on a modified convolutional neural network,” ETRI J., vol. 46, no. 4, pp. 633–647, 2024, doi: 10.4218/etrij.2023-0162.
[15] K. Chenglerayen, “From Reviews to Results?: Leveraging Amazon Feedback for Product Evolution From Reviews to Results?: Leveraging Amazon Feedback for Product Evolution,” no. December, 2024, doi: 10.32628/IJSRSET2411458.
[16] H. Boril, K. King, G. Strenski, A. Olson, and W. Husen, “Building Sentiment Analysis Pipeline: A Case Study On Amazon Reviews,” 27th IEEE/ACIS Int. Summer Conf. Softw. Eng. Artif. Intell. Netw. Parallel/Distributed Comput. SNPD 2024 - Proc., pp. 189–194, 2024, doi: 10.1109/SNPD61259.2024.10673954.
[17] M. S. Kumar, S. Bhagath, N. Tharun, K. Kishorekumar, and M. Narendar, “Transformer-Based Sentiment Analysis?: A Comparative Study of Machine Learning , Deep Learning , and BERT Models on Amazon Product Reviews,” vol. 2, no. 1, pp. 74–81, 2024.
[18] E. Hashmi and S. Y. Yayilgan, A robust hybrid approach with product context-aware learning and explainable AI for sentiment analysis in Amazon user reviews, no. 0123456789. Springer US, 2024. doi: 10.1007/s10660-024-09896-5.
[19] A. S. M. AlQahtani, “Product Sentiment Analysis for Amazon Reviews,” Int. J. Comput. Sci. Inf. Technol., vol. 13, no. 3, pp. 15–30, 2021, doi: 10.5121/ijcsit.2021.13302.
[20] N. M. Alharbi, N. S. Alghamdi, E. H. Alkhammash, and J. F. Al Amri, “Evaluation of Sentiment Analysis via Word Embedding and RNN Variants for Amazon Online Reviews,” Math. Probl. Eng., vol. 2021, 2021, doi: 10.1155/2021/5536560.
[21] M. V. Rao and C. Sindhu, “Detection of Sarcasm on Amazon Product Reviews using Machine Learning Algorithms under Sentiment Analysis,” 2021 Int. Conf. Wirel. Commun. Signal Process. Networking, WiSPNET 2021, no. October, pp. 196–199, 2021, doi: 10.1109/WiSPNET51692.2021.9419432.
[22] I. E. Fattoh, F. Kamal Alsheref, W. M. Ead, and A. M. Youssef, “Semantic Sentiment Classification for COVID-19 Tweets Using Universal Sentence Encoder,” Comput. Intell. Neurosci., vol. 2022, 2022, doi: 10.1155/2022/6354543.