Opinion mining in e-commerce: Evaluating machine learning approaches for sentiment analysis

In recent years, opinion mining has played a major role in analyzing text data from various sources such as Amazon, Capterra, Facebook, Google, GetApp, and Twitter. It enables companies to actively refine their business strategies. Sentiment analysis involves interpreting and classifying customer emotions (positive, neutral, and negative) expressed in reviews using sentiment analysis techniques such as BING and AFINN. This paper presents four approaches for customer review analysis and classification: the grade-based approach, content-based approach, content-based NRC-Emotion Lexicon approach, and collaborative approach. We employ three machine learning algorithms—stacking, random forest, and LogitBoost—to evaluate the performance of these approaches. A real-time dataset from Amazon product reviews is used for training and testing the model. Empirical results reveal that the collaborative approach outperforms the grade-based, content-based, and content-based NRC-Emotion Lexicon approaches across all three machine learning algorithms. Additionally, all approaches demonstrate outstanding performance when using the boosting algorithm for customer review classification.