Sentiment Analysis of E-commerce Product Reviews

Abstract

To monitor the reputation of companies, enhance the user experience, and recommend more personalized product offerings, online shopping websites increasingly depend on customer sentiment analysis. Sentiment analysis plays an important part in the context of user opinions and feedback, enabling companies to enhance services, products, and customer satisfaction. Nevertheless, current sentiment analysis systems suffer from a range of real-world issues, including the identification of sarcasm, processing multilingual reviews, and interpretation of context-dependent words. These issues lead to reduced accuracy and varying predictions. In this work, we present a hybrid sentiment classification approach that combines rule-based classifiers with BERT (Bidirectional Encoder Representations from Transformers) embeddings. The approach utilizes the contextual power of BERT and the precision of rule-based reasoning to identify the nuances in sentiment, especially in challenging or ambiguous reviews. We collected and preprocessed a customer review corpus of two of the largest online shopping platforms, Flipkart and Amazon, with web scrape techniques.

Text preprocessing included various natural language processing (NLP) methods like lemmatization, tokenization, and stopword removal to normalize and preprocess the text prior to passing it to machine learning models. Our hybrid model was compared against conventional models like Logistic Regression, Support Vector Machine (SVM), and Long Short-Term Memory (LSTM) networks.