ABSTRACT Purpose Livestreaming commerce is becoming increasingly frequent, and this rapid development has benefited from ever-increasing numbers of influencers joining businesses. Given this context, the purpose of this study is to examine the nonlinear effect of influencers’ beauty on viewer purchases. Methodology Based on the unstructured data of 2,597 livestreaming clips from TikTok, we first leverage a computer vision-based API interface, Face++, to calculate influencer beauty. Economic models are used to investigate this relationship. Second, the Sobel test is used to test the mediating effects of LSTM-based influencer trustworthiness. Finally, two moderators, product category and influencer type, are verified to clarify the boundary conditions. Findings First, there is an inverted U-shaped relation between influencer beauty and sales rate. Second, influencer trustworthiness mediates the impact of influencer beauty on sales rate. Finally, we find that compared with newcomer influencer selling experience products, veteran influencer beauty has a greater impact on sales when selling search products. Implications First, to the best of our knowledge, this is one of the few empirical studies evaluating influencer beauty through video data. More broadly, our research contributes to the rich influencer marketing literature by taking advantage of the abundant visual information provided by the new business context of livestreaming. The second is the new method of computer vision and multimodal machine learning. The combination of these new methods provides a scalable toolkit for unstructured data mining for academia and a realistic reference for livestreaming commerce for industry. The third contribution is content validity. We depend on hard archival data, which reflect changes in influencer beauty and viewers’ purchases in real time. This study extends the external validity of past studies and has greater reference value for real livestreaming businesses.
Read full abstract