Data-Driven Analysis of Usage-Feature Interactionsfor New Product Design

Abstract

Data-driven design has gained much attention with the rise of big data technologiesand the availability of user-generated data. Previous research utilizinguser data offered various design implications and ideas for new productdevelopment. However, most studies primarily focused on product featureswith little consideration of product usage, a significant factor in new productdesign. Moreover, while it is important in design practice to prioritize featuresin terms of usage, the interaction between usage and feature has rarelybeen investigated. To address the above limitation, this study proposes anew methodology that quantifies the interactions between product featuresand usages. The method consists of three stages. First, it analyzes customersentiments toward product features and usages. Second, the method trainsa neural network model that predicts customer satisfaction based on thesesentiments. Then the method calculates the impact of each input factorby SHapley Additive exPlanations (SHAP). In the final stage, the impactvalues are further analyzed by a new function, Effect Measurement basedon Covariance Analysis (EMCA), to quantify the interactions of feature andusage factors. The proposed methodology was initially tested on syntheticdatasets for validation and then applied to real-world datasets. The resultprovides numeric values for product usage-feature interaction, which can helpcompanies devise proper strategies for new product development.

Keywords: Data mining, explainable neural network, data-driven design, User experience, online reviews, sentiment analysis