Quantification and impact on sales of customers’ experience from reviews in social media

In recent years, conversations on social media have increased in number and intensity, making it possible to monitor and analyze consumer sentiment about brands, products, or services. This paper evaluates different machine learning and deep learning procedures to build proxy signals of consumer sentiment for improving product sales forecasts. We apply a multivariate autoregressive state space model with a bivariate observation constituted by the sales volume and a proxy of consumer sentiment and compare it with two univariate benchmarks not accounting for consumer sentiment.