Human-AI Alignment in Ad Targeting: Addressing Misestimation for Vulnerable Groups
研究微博女性用户对广告定向算法的误估（高估或低估），通过调查和深度访谈识别影响因素，提出理论模型，帮助广告商改进AI服务以避免信息茧房和不信任。
Users are playing an increasingly active role in receiving advertising information, making it essential to understand their assessment of algorithm from folk theory perspectives. In response to this need, we introduce the concept of “misestimation” to describe the gap between users’ perceptions of algorithmic systems and their actual operations. Focusing on the misestimation of targeting on Weibo, this research specifically focusing on female users’ overestimation/underestimation and the factors that contribute to these misestimations. Employing a mixed-method approach, the study conducted surveys and in-depth interviews with 25 participants. By identifying differences between overestimation and underestimation groups, the study developed a theoretical model of the themes influencing these perceptions: algorithm attitude, algorithm literacy, perceived surveillance, gender bias, and motivation to use platform. Methodologically, this study propose a mix-method to operationalize misestimation, which makes folk theory truly operationalizable. Theoretically, this article contributes a key concept for understanding human–machine alignment. Additionally, the findings advance the literature on uses and gratifications theory, algorithm attitude, and perceived surveillance theory by expanding them to human–machine alignment. By adjusting these influencing factors, this research offers suggestions to advertisers on Chinese platforms for how to improve AI service, in navigating personalized advertising content without falling into information cocoons or fostering distrust.