Extracting product design guidance from online reviews: An explainable neural network-based approach

Abstract

With the development of data mining techniques, user-generated data has become a valuable resource in diverse research areas. In product design research, many studies have been utilizing user data to discover implications for new product design. However, previous works focused on analyzing existing features, whereas companies also need strategies for new features. Some studies discovered new feature ideas from user data but did not provide design implications. This paper addresses the above limitation by extracting comprehensive design implications for both features from user data. The method first defines the lists of existing/new features and collects spec data for these features. Then, it constructs customer choice sets based on the online review and spec data. Regarding spec values, this study presents a newness merit function that reflects the changing value of new features and applies it to the choice sets. The final stage trains a neural network model based on choice sets and conducts SHAP (SHapley Additive exPlanations) on the model. The method draws design implications by further analyzing the resultant SHAP values. The suggested methodology was tested on real-world datasets. The result provides design guidance, including strategies for new features and recommended spec ranges for existing features. This article validates the result by showing that the obtained design implications are consistent with previous market research for product features.