A Comparative Study of Machine Learning Models for Sentiment Analysis: Customer Reviews of E-Commerce Platforms
Understanding customers' preferences can be vital for companies to improve customer satisfaction. Reviews of products and services written by customers and published on various online platforms offer tremendous potential to gain important insights about customers' opinions. Sentiment classification with various machine learning models has been of great interest to academia and practice for a while, however, the emergence of language transformer models brings forth new avenues of research. In this article, we compare the performance of traditional machine learning models and recently introduced transformer-based techniques on a dataset of customer reviews published on the Trustpilot platform. We found that transformer-based models outperform traditional models, and one can achieve over 98% accuracy. The best performing model shows the same excellent performance independently of the store considered. We also illustrate why it can be sometimes more reliable to use the sentiment polarity assigned by the machine learning model, rather than a numeric rating that is provided by the customer.
Davoodi, Laleh and Mezei, József, "A Comparative Study of Machine Learning Models for Sentiment Analysis: Customer Reviews of E-Commerce Platforms" (2022). Bled 2022 Proceedings. 27.