The power of talk: Exploring the effects of streamers' linguistic styles on sales performance in B2B livestreaming commerce

B2B livestreaming has become increasingly popular, but few studies have investigated the role of streamers in these B2B livestreaming events. In contrast to the existing studies that explored the appearance and professionalism of streamers and focused on the B2C livestreaming commerce, this study analyzes how streamers' linguistic characteristics relate to the livestreaming sales performance in online B2B marketplaces by conducting a streamer-level analysis. By combining deep learning and econometric analysis on a livestreaming dataset consisting of 915 live -streaming events across 14 industry categories from a B2B commerce platform, this study spe-cifically examines the impact of three speech acts (i.e., assertive, expressive, and directive speech acts) on sales performance. We first extract three speech acts by combining Word2Vec and Bi-GRU network with Attention Mechanism (WBA), then use econometric models to examine the effect of the streamer's speech acts on sales performance, and finally explore the moderating roles of product-level factors (i.e., product novelty and product category). The results show that the assertive and directive acts have a positive effect on sales performance, but the expressive acts have a negative effect on the sales performance. In addition, this study also provides strong ev-idence that the effects of these speech acts on sales performance are highly dependent on the novelty of the live product. This moderating effect is particularly significant for sensory products.