Visual materials have long played a pivotal role in shaping political narratives across different eras.Among the various forms of visual political content,portraits—or more broadly,human faces—hold a unique and significant position.As a core element of visual politics,the use of portraits is evolving in response to new technological contexts.The production and consumption of visual portraits are increasingly influenced by automation technologies,exemplified by the growing involvement of automated social media bots in the dissemination of portraits on various platforms.These changing dynamics necessitate a closer observation of the automation logic that now drives portrait politics and its impact on social media practices.To address this emerging landscape,this study focuses on Twitter(now renamed"X")to investigate the dissemination of portrait images by social media bot accounts from an algorithmic agent perspective.Using custom Python crawler scripts,the study collected 106,562 China-related images from September to November 2021.Social bot detection was then applied,revealing that 56,433 of these images—roughly 52.96％—were posted by bot accounts,while the remaining 50,129 images came from human users.To analyze the portrayal of human faces in these images,the study employed a computer vision tool to detect recognizable faces.Results indicated that 37,477 images(approximately 35.17％of the total)contained identifiable human faces,with 18,715 of these images originating from bot accounts.Leveraging the output of the computer vision analysis,the study further examined various facial attributes,including face size,position,gender,estimated age,facial expression,emotion,skin condition,blurriness,attractiveness,facial features,posture(such as open/closed eyes or mouth),and camera angles.The findings reveal that Twitter bots engage actively in the visual production of China-related content by selectively emphasizing certain demographic traits—such as gender and perceived age—as well as visual and facial features like expression,attractiveness,image size,position,and camera angles.Notably,bot-generated images differed significantly from those produced by human users.Combining these computer vision results with social media engagement metrics(likes and retweets),the study conducted negative binomial regression analysis,showing that bots manipulated the visibility of portraits to influence social media dissemination.Bots were more likely to use flat or bird's-eye views,depict more male portraits,show fewer smiles and more closed mouths(or masks),and post clearer images,all of which garnered higher engagement(likes and retweets),contributing to greater dissemination of these portraits.The study concludes that behind this manipulation of human faces by algorithmic agents in today's portrait politics lies a complex interplay of technological,platform,and political logics.At times,the technological and platform logics behind automated agents align directly with political objectives,while at other times,these logics diverge,with political priorities taking precedence.This convergence ultimately reshapes the landscape of visual politics in the digital age.
intelligent communicationinternational communicationcomputational communicationsocial media botportrait politicscomputer visionsocial media platformTwitter
万方数据
