Product Insights from Customer-Generated Data Using Topic Modeling with BERTopic and Sentiment Analysis with XLM-T: An Experiment on Turkish Reviews

Abstract

Abstract As information sharing through social media becomes widespread in every field, users frequently share their experiences with products purchased through e-commerce sites. This user-generated content is an opportunity for product owners to monitor users’ opinions. Since the number of user reviews is ever-increasing, decision makers need the right methods to monitor and extract valuable information from review data. In this study, an approach is proposed to determine the most prevalent product aspects and users’ opinions about them. In the proposed approach, first-topic modeling is applied to extract mostly debated product features. In addition, the monthly changes in the topics of reviews over time were examined using dynamic topic modeling. Next, sentiment analysis is applied to identify whether the customers like or dislike the features in the extracted topics. To apply the proposed approach, reviews about six similar security cameras were scraped from HepsiBurada.com, a famous e-commerce platform in Turkey. BERTopic is applied to extract topics, while XLM-T, a transformer-based technique, is implemented for sentiment analysis. From the experiments, fourteen topics related to product features were found. Extracted topics are mostly debated during the pandemic period. After the pandemic, the frequencies relatively stabilized. Among the extracted topics, ten have positive sentiment, while four have negative sentiment. The amount of review data is limited, and a different product with a large number of reviews can be used for further studies. Topics are manually named by exploring representative words. Further studies can utilize a tool for the automated labeling of topics using representative words. Sentiments about the product features may present valuable insight into product improvement scenarios. Additionally, the proposed approach can systematically identify product opportunities from a large amount of user-generated data.