Sales Prediction by Integrating the Heat and Sentiments of Product Dimensions

Abstract

Online word-of-mouth (eWOM) disseminated on social media contains a considerable amount of important information that can predict sales. However, the accuracy of sales prediction models using big data on eWOM is still unsatisfactory. We argue that eWOM contains the heat and sentiments of product dimensions, which can improve the accuracy of prediction models based on multiattribute attitude theory. In this paper, we propose a dynamic topic analysis (DTA) framework to extract the heat and sentiments of product dimensions from big data on eWOM. Ultimately, we propose an autoregressive heat-sentiment (ARHS) model that integrates the heat and sentiments of dimensions into the benchmark predictive model to forecast daily sales. We conduct an empirical study of the movie industry and confirm that the ARHS model is better than other models in predicting movie box-office revenues. The robustness check with regard to predicting opening-week revenues based on a back-propagation neural network also suggests that the heat and sentiments of dimensions can improve the accuracy of sales predictions when the machine-learning method is used.

Bicycle Sales Prediction Based on Ensemble Learning

“…In order to anticipate daily sales, they therefore suggested an autoregressive heat-sentiment (ARHS) model that incorporates the heat and sentiments of dimensions into the benchmark prediction model. Scholars undertake an empirical analysis of the film business and find that the ARHS model predicts box office receipts for movies more accurately than other models [4].…”