Understanding Consumer Opinions with Aspect-Based Sentiment Analysis: An Amazon Review Case Study

Finally, focusing on product features rather than overall sentiment, ABSA turns out to be a very important tool for the proper understanding of consumer opinions. We will apply ABSA to over 10,000 Amazon product reviews, extracting sentiment regarding quality, price, usability, etc., using both lexicon-based and machine learning-based approaches for performance in the accurate classification of specific sentiments for aspects. Our results show that deep learning models, especially BERT, achieved up to 88.7% in sentiment classification accuracy, compared with traditional machine learning models like Support Vector Machines (SVM) and Naive Bayes, with 78.3% and 75.6%, respectively. Although a bit slower, the lexicon-based methods showed off approximately 68.2% but are advantageous in obtaining quick, interpretable insights for real-time applications. Aspect extraction revealed that the reviews contained and mentioned price among other factors that the clients concerned most: 65% of the reviews mentioned the price, which was followed by quality at 58%, while durability was in 42%. Sentiments regarding durability appear to be very polarized; 40% of sentiments were very negative regarding durabilities. On the other hand, design received predominantly positive feedback in 70% of reviews. Results The results show that ABSA offers a fine-grained view of consumer feedback, and businesses can thus focus on specific points of the products they are improving. In future work, it would be possible to look at optimizing ABSA for different domains, so its applications may not be restricted to e-commerce.