Customer Review Analysis Using Word Embedding Model Considering Text Topics

Customers often give feedback on their evaluations and experiences with the products and service in the form of customer reviews, and developing the technology of customer review analysis plays an important role in the modern marketing research. Existing studies on marketing have used topic models to capture the review generating behaviors, but this approach ignores the word ordering, that is, it assumes a bag-of-words, and thus cannot adequately consider the context of text even with topic models. In this study, we propose a model combining supervised topic model and word embedding model for estimating the relationship between the product attributes mentioned in the customer review and their satisfactions while capturing the context of review text customers generate. In the empirical analysis, we apply the proposed model to a real customer review data on mascara-related products on a cosmetics e-commerce site, and the results show that our model captures some interpretable topics related to mascara products and estimates their effects on satisfaction scores, for example, the "eyelash" topic mentioned in the review tends to result in high levels of satisfaction, while the "brush" topic is associated with low levels of satisfaction.