Toward user-centered smart configuration systems for user-driven customization based on product usage context knowledge graph

Abstract

User-driven customization is a particular design paradigm that aims to fulfil individual customer needs while maintaining mass production efficiency. It has transformed the roles of customers from passive product recipients into active co-designers who can configure products that best meet their personal preferences and needs. With the technological advancements in artificial intelligence, smart configuration systems have been developed to improve co-design efficiency by proactively predicting customer needs and recommending design solutions. However, due to a lack of user-centered considerations, current smart configuration systems could undermine the core principles of customization in three significant ways. Firstly, existing systems still struggle to assist customers who have no prior product knowledge in performing the co-design process. Second, existing systems are heavily black-boxed and cannot provide customers with sufficient explanations about the reasoning behind design solutions. Third, due to limited data availability for minor market segments, current systems are prone to bias toward mainstream customers.

To address the above-mentioned challenges, this research firstly proposed a product usage context knowledge graph construction method using user-generated content, The proposed method can convert crowdsourced corner cases into a structured knowledge graph to support personal product usage context prediction, summarization, and reasoning. The product usage context knowledge graph can help customers anticipate the environmental and applicational factors that may affect product performance, assisting them in making more holistic decisions. Second, to address design biases in product usage context extraction, this thesis proposed a machine learning bias management framework. The framework will assist designers in fully considering and preventing the introduction of design biases when translating design problems into machine learning problems and developing machine learning models. Third, this thesis proposes a framework for smart configuration systems that integrate the proposed methods. With this system, customers only need to express their personal usage context to configure a product. The system also provides detailed explanations regarding the reasoning behind the configuration.