Customer Sentiment Analysis for Online Shopping Platforms Using NLP and Deep Learning Methods

Abstract

Customer sentiment classification has emerged as an essential part of digital commerce intelligence as online malls to shop in grow exponentially in product and user base. Social media like Flipkart have hundreds of product categories and millions of reviews. But the user feedback is very unstructured and may include HTML objects, URLs, punctuations and colloquialisms, which restrict the capability of contextual learning. In order to solve these difficulties, the structured Natural Language Processing pipeline is used, which involves removing null-texts, removing HTML tags and URLs, filtering by numbers and punctuation, removing stop-words, lexical-similarity stemming, tokenization, sentiment labeling by ratings, label encoding, and TF-IDF weighted features extraction. To be tested, the processed data is separated. It is advised to use a recurrent neural network (RNN) for categorization in order to preserve sequential sentiment and word order. The evaluation’s findings show that sentiment is very dependable, with 95.08% accuracy; SVM achieved 81.77% and 90.02% with a feedforward neural network. The rapid loss reduction, constant epoch convergence, and the limited validation deviation of 25 training cycles suggest that it can be easily used to ensure robust generalization when faced with class imbalance to enable scalable sentiment intelligence to be used in automated review interpretation, analytics, and adaptive decision support in e-commerce websites.