Importance-Induced Customer Segmentation Using Explainable Machine Learning

Abstract

Customer segmentation plays a critical role in enhancing a company's product penetration rate in the market. It enables numerous downstream applications such as customer-oriented product development and trend analysis. Previous approaches to customer segmentation have relied either on survey-based methods or data-driven approaches. However, these methods face challenges such as high human labor requirements or the generation of noisy segments. To address these challenges, this paper proposes a new methodology based on data-driven network construction and an importance-enhanced framework. The framework incorporates two techniques: (1) the utilization of a neural network model to compute feature importance values and (2) the proposal of a novel network connection rule. This framework addresses the limitation of the previous approach, sentiment-polarity-based networking, by connecting customers based on feature importance. We further validated the effectiveness of the framework using three real-world datasets and demonstrated that the proposed method outperformed the previous approach.