How uncanny are virtual vs. human influencers: A text mining approach

Abstract

This paper uses the uncanny valley theory to examine how users react to different development levels of human-like avatars on social media. Three virtual avatars with different human-like appearances are compared with a human influencer from the same peer group on Instagram. The methodology is based on the evaluation of user comments on the posted content of the subjects. Text mining is used to evaluate the comments and calculate a sentiment score for each influencer. The results indicate that users react positively to the human influencer and negatively to the most human-like avatar. The study shows users follow the uncanny valley curve when dealing with virtual influencers. When users deal with virtual influencers who try to imitate humans but have low CGI quality, the results are positive. However, when virtual influencers are close to human-like appearance, there are adverse effects on their satisfaction.