Over the past few decades, there has been an increasing emphasis on a company’s ability to produce high-quality consumer products. Such products can be identified by measuring the associated customer satisfaction level. Therefore, this emphasis has gradually transformed most industries from production-centralized to customer-driven ones. Market analysis is an effective means to understand customer perception towards new consumer products. Data collection tools such as questionnaires and users’ interviews, can be used in this regard. Based on the survey data, customer satisfaction models developed can be used to identify customer perceptions towards new products and the associated customer satisfaction level. Customer satisfaction has a direct influence on customer retention (Choi et al., 2004 and Hansemark and Albinsson, 2004) and company’s profitability (Johnson et al., 1996 and Zeithaml, 2000). In this regard, it is crucial to improve customer satisfaction and identify the associated design attributes that would ensure sustained customer loyalty and competitiveness for the firm (Deng & Pei, in press).
Previous studies have attempted to develop customer satisfaction models with statistical regression, fuzzy regression, neural networks, quantification analysis I, and fuzzy rule-based modeling. Chen, Khoo, and Yan (2006) developed a prototype system for affective design in which Kohonen’s self-organizing map neural network was employed to consolidate the relationship between design attributes and customer satisfaction. Hsiao and Tsai (2005) proposed a method that enables an automatic product form search or product image evaluation by means of a neural network-based fuzzy reasoning genetic algorithm. The neural network-based fuzzy reasoning algorithm was applied to establish relationships between the input form parameters and a series of adjectival image words. Fung, Popplewell, and Xie (1998) proposed fuzzy rule-based models to relate design attributes to customer satisfaction. Han, Yun, Kim, and Kwahk (2000) developed a variety of usability dimensions, including both subjective and objective aspects, and evaluated product usability based on statistically regressed models. These models were then used to identify functional relationships between design attributes and customer satisfaction. At the same time, various techniques have been attempted to model the fuzzy relationships between design attributes and customer satisfaction. Kim and Park (1998) suggested a fuzzy regression approach to estimate functional relationships. Chen, Tang, Fung, and Ren (2004) proposed another fuzzy regression approach, based on asymmetric triangular fuzzy coefficients, to model the functional relationships. The use of non-linear programming to develop fuzzy regression models for the functional relationships was proposed by Chen and Chen (2005). However, the above approaches are only applicable to developing linear models, and ignore non-linear terms of models. Multiple linear regression, which considers non-linear coefficients, was attempted to model the relationships between customer requirements and engineering characteristics (Dawson & Askin, 1999). However, an optimal model could not be generated because the model is in a polynomial form, and the order of the polynomials generated is user defined. Liu, Zeng, Xu, and Koehl (2008) proposed a fuzzy model to examine the customer satisfaction index in e-commerce. They considered a method that would calculate the index based on a 5-level quantity table using fuzzy logic. However, the developed model is implicit, in other words, a black-box model. Grigoroudis and Siskos (2002) developed the MUlticriteria Satisfaction Analysis (MUSA) method for measuring and analyzing customer satisfaction. MUSA is a preference disaggregation model based on the working principles of ordinal regression analysis. Using the survey data, MUSA aggregated individual judgments into a collective value function for quantifying customer satisfaction. The model assumed that the overall customer satisfaction was measured solely with respect to several customer attributes, and ignored the customer satisfaction model towards each customer attribute. Grigoroudis, Litos, Moustakis, Politis, and Tsironis (2008) further applied the MUSA method to measure the user-perceived web quality. Park and Han (2004) proposed a fuzzy rule-based approach to examine customer satisfaction levels towards office chair designs. They reported that the fuzzy rule-based approach outperformed the multiple linear regression approach in terms of the number of variables used. Similarly, Lin, Lai, and Yeh (2007) proposed a fuzzy logic model to determine the consumer-oriented mobile phone form design. The experimental results suggest that the fuzzy model outperformed two neural network-based models in terms of the root of mean square errors. You, Ryu, Oh, Yun, and Kim (2006) developed the customer satisfaction models for automotive interior material using quantification I analysis, which examined the relatively important design variables and preferred design features. Hence, significant design variables and their values affecting customer satisfaction were identified. However, the models developed are implicit.
Before endeavouring to develop customer satisfaction models, the vague affiliation between customer attributes and design attributes must be thoroughly investigated (Kim et al., 2000 and Kwong et al., 2007). The literature review indicates that most of the existing models assume a linear relationship between customer attributes and design attributes. In fact, they could be highly non-linear. Although neural networks are capable of modeling the non-linear relationships between them, the customer satisfaction models generated are implicit. Thus, it is difficult to analyze the behaviour of the relationships. In this paper, a new methodology for developing customer satisfaction models using a neuro-fuzzy approach is proposed, whereby non-linear and explicit customer satisfaction models can be generated.
The organization of this paper is as follows: Section 2 describes the main steps of the proposed methodology. Section 3 presents an illustrative example to demonstrate the usefulness of the methodology. Two validation tests are depicted in Section 4. Conclusions are given, together with future research work, in Section 5.
