Supporting new product development using customers’ online data and computational intelligence methods

Abstract

In recent times, the availability, and the proliferation of the generation of online data derived from social media and e-commerce platforms have been capitalised by firms in different ways to influence, promote, enhance, and develop new products. In the area of the new product development (NPD) process, online data can be applied differently in each stage of the NPD process. Within the NPD process, enhancing customer satisfaction and making product demand forecasts are two areas that require the extensive use of data. Previously, to obtain data for NPD, surveys were mostly conducted by product manufacturers to seek information's from customers before designing a new or improving a new product. However, the nature of conducting surveys tends to be cumbersome and respondents can easily misinterpret the questionnaires. Surveys also have the limitation of being incomplete and, usually, the ratings used in surveys do not convey the real needs of respondents. Thus, developing customer satisfaction models from surveys presents many complexities since customers' responses' fuzziness is usually not considered. Similarly, the identification and predicting of the most important product attributes have not been explored in past studies in addressing the dynamic needs of consumers. This is due to the over reliance on surveys that fails to provide reliable data for manufacturers, thus preventing them from producing products that meet the rapid changes in customer needs due to technological advancement. As part of product development activities, the demand for products is usually forecasted to prevent revenue loss. However, most of these forecasts require large amount of historical data to develop a demand forecast model. With the advent of the internet, manufacturers can integrate constantly updated user generated online data in forecasting models in order to forecast the adoption of products. To overcome the above limitations, the objectives of this research are presented in three phases: i) To propose a novel customers satisfaction model that address the fuzziness and nonlinearity of customer satisfaction models using multigene genetic programming based fuzzy regression (MGGP-FR) ii) To formulate a methodology for determining and predicting the importance of product attributes. The Shapely Value and Choquet integral are employed to estimate the importance of product attributes and based on the importance values, a fuzzy rough set times series method is proposed to forecast the future importance of product attributes. iii) To propose a new market share model and demand forecasting model that addresses uncertainties in forecasting. A market share model is developed from the multinomial logit (MNL) model and the fuzzy regression (FR) approach while the demand model is developed from a modified Bass model integrated with sentiment scores from online reviews. A case study on modelling customer satisfaction for electronic hairdryers using MGGP-FR is presented in this study. To validate the proposed methodology, the results of the MGGP-FR are compared with previously proposed methods mainly FR, genetic programming (GP), and genetic programming-based fuzzy regression (GP-FR). Based on the mean relative errors and the variance of errors of the MGGP-FR and previous methods, the proposed MGGP-FR showed a better performance when compared with the previous methods. Next, forecasting the future importance of the product attributes of an electronic hairdryer is illustrated using the fuzzy rough set time series method. The proposed fuzzy rough set time series forecasting accuracy outperformed the fuzzy time series method. Lastly, a case study on forecasting the adoption of a Tablet P.C is used to illustrate the applicability of the proposed fuzzy modelling and discrete choice analysis method for forecasting product adoption using online reviews. The proposed method was compared with the fuzzy time series forecasting and the original Bass model and was found to be better as it provided different scenarios for the forecast and acceptable forecasting results.