Improving the Bass model’s predictive power through online reviews, search traffic and macroeconomic data

Abstract

The Bass diffusion model has been successfully applied in product sales forecasting, and it performs particularly well in consumer durables sales forecasting. However, the traditional Bass model only uses historical sales data and cannot contain important market information concerning products. How to improve the Bass model through user-generated Internet information and macroeconomic data to achieve more accurate predictions is addressed in this paper. First, a sentiment analysis is adopted to convert online reviews concerning various attributes of a product into sentiment scores, and then, the product word-of-mouth index (WoM index), which is integrated into the imitation coefficient of the Bass model, is calculated by the entropy weight method. Subsequently, the Baidu product index is calculated through the Baidu search traffic of product-related words and is integrated into the innovation coefficient of the Bass model. Finally, macroeconomic data are collected to estimate the total number of potential adopters, which relaxes the assumption that the market potential in the Bass model remains unchanged over time. We conduct comparison experiments of forecasting automobile product sales, and the results are as follows. (1) The improved Bass model can significantly improve the forecast accuracy, and its average forecast accuracy (0.9983) is approximately 2.15% higher than the traditional Bass model (0.9773). The RMSE (0.3124) and WAPE (0.0017) are 90.98% and 92.38% lower compared with the traditional Bass model, respectively. (2) as for the calculation of WoM index, it is better to divide a whole review into separate reviews concerning each attribute. (3) Macroeconomic data play the biggest role in improving the prediction power of the Bass model, followed by online review data and search traffic data.