Comparison of LDA, NMF and BERTopic Topic Modeling Techniques on Amazon Product Review Dataset: A Case Study

With the developing technology, the e-commerce market is growing day by day. As of 2022, it is estimated that 19.7% of the sales in the world are made over the internet. However, there are negative elements that distinguish online sales from regular sales. Communication between the seller and the customer is more difficult on the online platform. Likewise, problems such as quality or cargo are constantly written under the product reviews. For this reason, the seller must constantly monitor customer feedback and take the necessary action. With topic modeling algorithms, user complaints can be grouped and read in groups. In this study, LDA (Latent Dirichlet allocation), NMF (Non-Negative Matrix Factorization) and BERTopic algorithms tested on Amazon product review dataset were compared. According to the results obtained, all 3 algorithms are successful and useful. The BERTopic algorithm produced more meaningful results than other algorithms according to the consistency calculation metric.