Extraction of customer satisfaction topics regarding product delivery using non-negative matrix factorization

The purpose of this research is to extract knowledge about textual data of customer satisfaction from the customer's voices with regard to delivery of e-commerce products. With the recent development of e-commerce, new distribution formats such as O2O and Omni-Channel are penetrating society. Since customer reviews are free-form descriptions, comments about delivery services and comments about the personal experience of individual customers are mixed in with product evaluation data. In this study, we focus on the topic extraction function of nonnegative matrix factorization and extract topics from mixed review data. We propose a method of decomposing review data and extracting topics to identify delivery-related expressions. Furthermore, by using a random forest with the above topic as the features, we were able to detect those factors affecting overall satisfaction.