Influencing factors of livestream selling of fresh food based on a push-pull model: A two-stage approach combining structural equation modeling (SEM) and artificial neural network (ANN)

Abstract

Under the COVID-19, fresh food e-commerce has acquired new sales channels through live shopping, and the use of live broadcasts has become a hot spot in management and practice. However, there is little empirical evidence of the influence of live streaming on sales. This study combines the perspective of People-Goods-Scene and the push-pull theory, and proposes a two-stage method for forecasting sales volumes using structural equation models and artificial neural networks. It was found that the number of page views was the strongest predictor of live broadcast sales, while the numbers of interactive comments, live broadcasts with goods, and videos with goods, together with clean labels were weakly predictive. A comprehensive neural network model showed an accuracy of 83.76% in the prediction of live broadcast sales. These research results provide a theoretical basis for the prediction of fresh food shopping behavior in live-broadcast e-commerce from the perspectives of the consumers and the goods yard and provide ideas for the design of live broadcast content and optimization of user experience.