In a person’s hectic schedule, online buying is a more convenient option. However, this retail center is teeming with fictitious products, fake brands, and low-quality goods. It is the job of shopping sites to employ some kind of sophisticated algorithm to assess the popularity, suggestion, and quality of each accessible product and brand for customers. These shopping sites collect descriptive reviews or comments from users, and sentiment analysis is performed to determine the quality, popularity, and adaptability of these products. A client submits text-based reviews to share his thoughts, opinions, and feelings about the advantage or loss he experienced with that deal. On a shopping or emarket site, thousands of reviews are given by customers, making it impossible to assess each review individually. Such sites can apply a customer satisfaction analysis algorithm to exactly measure the evaluations provided. Using this customer satisfaction research, we may determine whether the product was favorably received by the buyer. However, conventional sentiment analysis methods struggle when a single feeling is combined with many sentiments of distinct elements. For evaluating feelings, an intelligent aspect-based and weighted assessment adaptive CWA-wbiLSTM Model is proposed in this study. The processing featureset in this model is enhanced by combining the BoW and Handcrafted weighted features. The aspect-oriented clustering stage categorizes and strengthens this extended featureset based on aspects and weights. For classifying the reviews, an updated aspect-oriented and weighted LSTM model is implemented in the final stage. The proposed methodology has been tested on Amazon product reviews for cell phones, video games, and electronics. The models being compared include SVM, LSTM, CNN, RNN, biLSTM, Tree + LSTM, and LSTM + FL. The proposed model was 96.15% accurate for Amazon Cell Reviews, 97.98% accurate for Amazon Video Games, and 97.42% accurate for Electronics Products. SVM, CNN, LSTM, RNN, BiLSTM, Tree-LSTM, and LSTM + FL models yield significant increases in an average accuracy of 24.91%, 41.54%, 7.67%, 8.02%, 7.74%, 7.39%, and 2.16%. Current SVM, LSTM, RNN, BiLSTM, Tree-LSTM, and LSTM + FL models have an average accuracy of 72.56%, 88.93%, 88.91%, 89.10%, 89.12%, and 94.16%, respectively. The suggested CWA-wbiLSTM model surpassed all of the prior models, with an accuracy of 97.98%. The proposed model achieved a fantastic accuracy of 96.92% for quality, 96.26% for look, 96.56% for price, 99.31% for color, and 96.90% for size. The findings corroborated each measure’s genuine precision rate. With maximum precision and recall rates of 0.981 and 0.98, respectively, the suggested CWA-wbiLSTM model surpassed both baseline machine and deep learning models. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.
