Topic sentiment mining for sales performance prediction in e-commerce

Abstract

In the era of big data, huge number of product reviews has been posted to online social media. Accordingly, mining consumers’ sentiments about products can generate valuable business intelligence for enhancing management’s decision-making. The main contribution of our research is the design of a novel methodology that extracts consumers’ sentiments over topics of product reviews (i.e., product aspects) to enhance sales predicting performance. In particular, consumers’ daily sentiments embedded in the online reviews over latent topics are extracted through the joint sentiment topic model. Finally, the sentiment distributions together with other quantitative features are applied to predict sales volume of the following period. Based on a case study conducted in one the largest e-commerce companies in China, our empirical tests show that sentiments over topics together with other quantitative features can more accurately predict sales volume when compared with using quantitative features alone.