Interpretable Perceived Topics in Online Customer Reviews for Product Satisfaction and Expectation

Online customer reviews contain useful and important information, in particular, for product management--because customers tend to praise or criticize certain features or attributes of goods in their reviews. We propose a model that extracts the perceived topics from textual reviews by natural language processing under the restrictions of their interpretability and predictability of product satisfaction as current product evaluation and expectation as future possible demand by supervised learning. The empirical analysis on user-generated content of food reviews shows that our proposed model performs better than alternative models, and it suggests product managers the necessity of improving some specific attributes and focus their advertising on these attributes as fulfilling customer needs.