Mass Customization (MC) is a production strategy aiming at providing goods and services to meet individual customer’s needs with near mass production efficiency [1]. It is considered as the key to the economic growth in the 21^st century [2]. Product configurator (PC) has been widely recognized as the effective tool to generate MC orders [3]. However, current PCs often require customers to make choices based on the product design parameters which they are unfamiliar with, rather than based on their actual perceptual needs. This difference of scopes can be characterized as the ‘semantic gap’, meaning the discrepancy between two representations of the same object. It increases the confusion and difficulty that customers encounter in configuration, which is a major challenge for MC [4]. To...[ Read more ]

Mass Customization (MC) is a production strategy aiming at providing goods and services to meet individual customer’s needs with near mass production efficiency [1]. It is considered as the key to the economic growth in the 21^st century [2]. Product configurator (PC) has been widely recognized as the effective tool to generate MC orders [3]. However, current PCs often require customers to make choices based on the product design parameters which they are unfamiliar with, rather than based on their actual perceptual needs. This difference of scopes can be characterized as the ‘semantic gap’, meaning the discrepancy between two representations of the same object. It increases the confusion and difficulty that customers encounter in configuration, which is a major challenge for MC [4]. To improve the performance of PCs and achieve the aim of MC, the relationship between the customer needs and the design parameters is studied to bridge the semantic gap.

In 2007, Randall et al. introduced the concept of Needs-Based Product Configurator [5] to bridge this semantic gap in PCs, which allows customers to directly configure orders with their perceived needs. It is demonstrated that this configurator can result in better performance in terms of customer satisfaction towards the product and shopping process than traditional PCs for innovative products. However, there are several critical limitations hindering it from implementation in real business. Firstly, the embedded relationship in the PC between the customer needs and the product parameters is predefined by the designers, which can be biased and is not reliable for general products and subjective attributes. Also, this method is not feasible for a large product solution space and it is difficult to be responsive for any market change. Moreover, their design only allows the pure needs-scope configuration, which is still not natural for customers. Therefore, a more practical and intelligent design methodology is needed to directly retrieving the semantic mapping information from the customers and to overcome the above limitations.

In this study, Relative Attribute Recognition, one of the most up-to-date methodologies for learning human semantics and bridging semantic gap in computer vision [6], is applied to generate customer perceptional descriptions on product alternatives. This method considers every customer perceived attribute in the relative comparison (e.g. this is an object larger/smaller than the reference) instead of the binary existence (e.g. this is a large/small object), which is proved to be more robust for the customer subjectivity [6]. We then build a generative model to map between the physical features and customer perceptions (i.e. bridge the sematic gap in PCs). Lastly, we propose a framework of implementing Needs-Based Configurator for general configurable scopes and product domains.

Experiments are conducted to verify that our proposed design can overcome the mentioned limitations of the previous Needs-Based Product Configurator. The proposed PC can be extended for general products and it indeed results in better performance than the traditional PCs. As result, customer perceived satisfaction towards the MC offerings has been significantly improved. Some future directions of improving the configurator are also inspired by cross-checking with the applications in the computer vision papers where our applied methodology originates, and discussions are conducted accordingly.