A Domain Knowledge Integrated Convolutional Neural Network for Translating Customer Needs Into Configuration Choices in Mass Customization

Abstract

Mass customization has emerged as a viable smart manufacturing strategy to deliver tailor-made products with the efficiency of mass production. It significantly impacts a company’s research, development, and engineering functions by fostering innovation in product design, manufacturing processes, and supply chain management. A critical challenge in mass customization is developing a user-friendly choice navigation process that enables customers to identify customized designs with minimal burden and complexity. This article addresses this challenge by proposing a novel approach to choice navigation that maps customer needs expressed in natural language to suitable product attribute choices. We tackle data sparsity issues by leveraging the extensive amount of online product-review text to mine customer needs and preferences. External domain knowledge in the product domain is distilled using conceptual graphs. We then develop a convolutional neural network-based structure and a transfer learning procedure to integrate this domain knowledge with contextual semantic information from the review and needs text. Our extensive experiments show that the approach’s effectiveness and robustness in the needs-attributes mapping, and demonstrate its potential to improve user-friendliness and customer satisfaction in mass customization systems.