Topic Modeling-Based Framework for Extracting Marketing Information From E-Commerce Reviews

Abstract

Reviews left by consumers on e-commerce platforms provide crucial marketing information as they are a publicly available source of information providing insight into consumers’ thoughts and opinions. However, it is physically impossible to read hundreds of reviews per product. Therefore, algorithmically extracting the necessary insights from a large volume of review data can provide more advanced information while reducing time and costs. This study aims to automate the process of extracting related products, pros and cons of products, and trend forecasting from review data using clustering algorithms. The review dataset for pros and cons of products and related products was constructed by selecting 17 products on the Naver Shopping platform and crawling them and the product keyword and search volume dataset for trend forecasting was constructed by crawling the data from Itemscout and Naver datalab platform. Various clustering-based algorithms such as Deep Clustering Network, BERTopic, and a Transformer-based forecasting model were used to conduct the research. It is expected that this will allow for a more accurate understanding of consumer thoughts, which can be utilized in marketing for various products and services.

Keywords

• Topic modeling
• clustering algorithm
• trend forecasting
• BERT
• e-commerce marketing