A two-phased SEM-neural network approach for consumer preference analysis
Introduction
Consumer preference analysis is a fundamental task in the design process of consumer products. The primary focus of this task is establishing a mapping relationship between consumer preferences and product parameters/attributes. The key to connect the consumer space and the design space are user perceptions of the product. User perception refers to the unique experience that users have when using a product. Users usually use words like comfort, satisfactory, attractive, slow, unreliable to describe the experience. For the fact that the preference model reveals the relationship between the consumer preferences and product attributes, it is widely used in product development, market segmentation, brand competition analysis, pricing strategy and other fields.
Traditionally, methods like choice analysis (CA) are used to map the consumer information onto a single construct—utility. The consumer preferences are determined by the utility provided to the consumer by a product or service. The utility refers to the ability of a product or service to satisfy the user’s needs and desires, which reflects the satisfaction that consumer felt when using a product or enjoying a service. The Discrete Choice Models (DCM) is a common method for consumer preference analysis [1], [2], [3], [4]. On the basis of DOE (Design of Experiment) [5], the DCM can measure user's purchase behavior by simulating the product or service in the market competition environment. The user’s choice between different products or services can be obtained through the DCM. In DCM, the dependent variables are multiple products that can be selected by users and the independent variables are the different product attributes, so nonlinear multivariate regression is a main analysis method for DCM. The logistic regression was used to analyze the relationship between dimensions of international express service and user preference, so that to provide new ideas for the design of international express service [6]. The multiple linear regression and support vector regression was used to analyze the relationship between product design features and user preference, which can help designers optimize product design and improve practical values [7]. Besides, other methods to analyze the user preference in the field of economics include correlation analysis [8], clustering analysis [9], neural network [10], network analysis approach [11], [12] and so on. For all above methods that originated in the market domain, they allow designers to understand the important design parameters/attributes that affecting the fulfillment of consumer perceptions and ultimately their preferences [13]. In these methods, however, the user information is usually mapped to a set of utilities. The designers can only get what the user’s preference is, but do not know how the specific preference is generated or which attribute has more influence on the user’s preference. That means these methods cannot reveal the specific emotional changes of users in the mapping procedure.
Another dimension to study the consumer preference comes from the behavioral research in the field of psychology. Theory of Reasoned Action (TRA) [14] expatiates people are rational individuals and they would consider the effects and results before taking any action. So TRA can be used to analyze the determinants of conscious intention and behavior. The premise of TRA is that individuals have rational control abilities, but, the actual behavior cannot be determined only by the subjective factors of the individual, but also influenced and restricted by the objective factors.
In the study of technology adoption/acceptance model, Davis [15] proposed the Technology Acceptance Model (TAM), trying to explain the impact of external factors, user beliefs and user attitudes on the usage intention and actual usage behavior of the information system. TAM has been widely used in interpreting and predicting user information technology acceptance behavior, and some improved models of TAM have been developed to analyze user's behavior in other fields, such as TAM2 (Technology Acceptance Model2) [16] and UTAUT (Unified Theory of Acceptance and Use of Technology) [17] and TAM3 (Technology Acceptance Model3) [18]. In above methods, user's ideas, perceptions, and attitudes are mapped into a network to predict downstream user preference. Through this network the reasons behind users’ specific perception or preference can be traced, which can provide designers with more abundant information about user preference.
In order to verify the relationship between the factors of network, the Structural Equation Model (SEM) [19] is a commonly used technique for parameter estimation and hypothesis testing. SEM is a general statistical technique for the estimation of a system of simultaneous linear equations that may include both observed and latent variables [20]. That is, SEM techniques have made it possible for researchers to examine theory and measures simultaneously [21]. SEM are widely used in examining a variety of structures, including causal models, measurement models, growth models, and combinations of these. For instance, in Ref. [22], SEM was used to explore how the performance of the project participants affects the contractor's satisfaction. They established a model to analyze the customer's goal clarity, construction risk management, and mutual respect and trust on the impact of the customers’ satisfaction. Ghosh et al [13] mapped the user’s various psychological constructs and their relationships into a network, and then used the SEM to verify the proposed model. Despite that the SEM method has been widely applied in many fields, the assumption that the relationships between variables are linear limits the depth of its application. Previous research have proved that there is a nonlinear relationship between product quality and customer satisfaction [23], [24]. In some cases, it may oversimplify the complexities involved in the causal paths.
With the increasing complexity of the model, researchers have called for new SEM techniques to solve this issue. Some scholars try to improve the conventional SEM to express the nonlinear relationship between variables by introducing the quadratic and interaction term of the variables into the model [25], [26]. However, one substantial drawback is that such approaches assume that the latent linear predictor variables are multivariate normally distributed. When this assumption is violated, the parameter estimates in the nonlinear SEM can be biased [27]. Fortunately, the Artificial Neural Network (ANN) [28] provides the ability of self-learning and the nonlinear characteristic to fit the nonlinear relationship between multiple variables. In general, the topology of an ANN is a fully connected network that determined by the experience of engineers, which makes it difficult to explain the causal relationship between input and output variables. In addition, it is often difficult to construct a neural network model when learning from data. Employing the results from SEM provides a way to develop a neural network model with a good prediction performance [29]. There are different types of multiple-analytic approaches that combined SEM and ANN. For instance, SEM can be used to analyze part of the question responses, while the rest of the responses can be predicted by ANN [30]. Hackle and Westlund [31] contended that the ANN-based SEM technique can be superior to conventional SEM because it can measure nonlinear relations by using different activity functions and layers of hidden nodes. Scott and Walczak [32] applied a multi-analytic approach for assessing computer self-efficacy (CSE) and technology acceptance, in which the SEM were used to test the hypotheses and the reliability of the measures, while NN analysis verifies the antecedents as predictors of CSE, estimates CSE scores, and assigns individuals to groups based on their CSE scores. Using the output result of ANN as SEM’s input is also a common approach [32], [33]. Although the SEM-neural networks have been applied to some problems, including assessing technology acceptance [29], [33], [34], [35] and customer relationship management (CRM) adoption [36], they have not been previously used for consumer preference analysis.
In this paper, a two-phased SEM-neural network approach is proposed, in which the topological structure is obtained from the conventional SEM and the path coefficients are obtained from the BP algorithm-based training process. Firstly, the causal relationship between the product attributes and user preference are obtained from the topological structure of the conventional SEM. Then, the self-leaning process based on the BP algorithm is conducted to train the ANN based on SEM results to get the revised path coefficients. Finally, the proposed model is used to fit the user’s preference on the smartphone in a case study.
Section snippets
Methodology
The proposed method includes two phases.
In Phase I, the SEM is used as a parameter estimation and hypothesis testing technique to obtain the influence path between the user’s multi-dimensional perceived performances and user’s preference. Firstly, the assumptions of the user preference model are proposed, including: 1) the relationships between the dimensions of the perceived performances and the user preferences; 2) the indicators for each dimension of the perceived performances. Then, a
Case study
Smartphone is one of the most representative smart products. And the smartphone manufacturers have paid much attention to the perceived performances of smartphone users. A case study of smartphone is presented in this section to demonstrate the effectiveness of the proposed methodology.
Conclusions and future work
In the design process of consumer products, a primary task is establishing a mapping relationship between consumer preferences and product attributes. The proposed SEM-NN approach can be useful to help the designers to identify and map how product attributes affecting the fulfillment of user perceptions and ultimately their preferences, and to better understand the factors that affect user perceptions and the inner relationships between them. The SEM-NN model can make full use of the causal
Funding
This research is supported by the National Natural Science Foundation of China (Grant Nos. 51875345, 51475290).
Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
References (49)
- et al.
Market Demand Estimation for New Product Development by Using Fuzzy Modeling and Discrete Choice Analysis
Neurocomputing
(2014) - et al.
Discrete Choice Models with Latent Choice Sets
Int. J. Res. Market.
(1995) - et al.
Preference-Based Clustering Reviews for Augmenting E-commerce Recommendation
Knowl. Base. Syst.
(2013) - et al.
Examining the Influence of Participant Performance Factors on Contractor Satisfaction: A Structural Equation Model
Int. J. Project. Manage.
(2014) - et al.
A SEM–Neural Network Approach for Understanding Determinants of Interorganizational System Standard Adoption and Performances
Decis. Support. Syst.
(2012) - et al.
Predicting Consumer Decisions to Adopt Mobile Commerce: Cross Country Empirical Examination between China and Malaysia
Decis. Support. Syst.
(2012) - et al.
Cognitive Engagement with A Multimedia ERP Training Tool: Assessing Computer Self-Efficacy and Technology Acceptance
Inform. Manag.
(2009) A Two-Staged SEM-Neural Network Approach for Understanding and Predicting the Determinants of M-Commerce Adoption
Expert. Syst. Appl.
(2013)- et al.
Predicting the Drivers of Behavioral Intention to Use Mobile Learning: A Hybrid SEM-Neural Networks Approach
Comput. Hum. Behav.
(2014) - et al.
Forecasting Social CRM Adoption in SEMs: A Combined SEM-Neural Network Method
Comput. Hum. Behav.
(2017)