Sentiment Classification of Product Reviews using Machine Learning

In the fast-paced e-commerce environment, understanding and learning consumer perceptions of products is essential for businesses to enhance user experience, optimize marketing strategies, and better overall customer satisfaction. This paper investigates the use of machine learning models for sentiment classification and sentiment analysis. VADER, known for its simplicity and speed, employs a sentiment lexicon for polarity scoring. RoBERTa and BERT transformer-based models, which excel in capturing contextual information and long-range dependencies, making them extremely powerful for sentiment analysis. The performance of rules-based models typically falls below Machine Learning models, which on the other hand, generally exhibit better performance. A diverse dataset is used and the preprocessing steps are detailed. This includes duplicate removal, training of models, using NLP etc. In this paper the highest F1 score is obtained in Bert is 68.9 followed by a 57.8 in RoBERTa and then Vader with a 12.7. These show the superior capability of machine learning models like BERT and RoBERTa in capturing sentiment compared to rule based approaches like Vader which is simpler in nature. This shows the superiority of these models over older models like SVM and Naive Bayes often showing lower average F1 scores. This paper contributes valuable guidance for implementing effective sentiment analysis systems in E-commerce.