Langner Moritz, Toreini Peyman, Maedche Alexander
Institute for Information Systems (WIN), Department of Economics and Management, Karlsruhe Institute of Technology (KIT), Kaiserstraße 89-93, 76133 Karlsruhe,
J Eye Mov Res. 2025 Apr 1;18(2):8. doi: 10.3390/jemr18020008. eCollection 2025 Apr.
Eye-tracking technology provides high-resolution information about a user's visual behavior and interests. Combined with advances in machine learning, it has become possible to recognize user traits and states using eye-tracking data. Despite increasing research interest, a comprehensive systematic review of eye-based recognition approaches has been lacking. This study aimed to fill this gap by systematically reviewing and synthesizing the existing literature on the machine-learning-based recognition of user traits and states using eye-tracking data following PRISMA 2020 guidelines. The inclusion criteria focused on studies that applied eye-tracking data to recognize user traits and states with machine learning or deep learning approaches. Searches were performed in the ACM Digital Library and IEEE Xplore and the found studies were assessed for the risk of bias using standard methodological criteria. The data synthesis included a conceptual framework that covered the task, context, technology and data processing, and recognition targets. A total of 90 studies were included that encompassed a variety of tasks (e.g., visual, driving, learning) and contexts (e.g., computer screen, simulator, wild). The recognition targets included cognitive and affective states (e.g., emotions, cognitive workload) and user traits (e.g., personality, working memory). A set of various machine learning techniques, such as Support Vector Machines (SVMs), Random Forests, and deep learning models were applied to recognize user states and traits. This review identified state-of-the-art approaches and gaps, which highlighted the need for building up best practices, larger-scale datasets, and diversifying tasks and contexts. Future research should focus on improving the ecological validity, multi-modal approaches for robust user modeling, and developing gaze-adaptive systems.
眼动追踪技术可提供有关用户视觉行为和兴趣的高分辨率信息。结合机器学习的进展，利用眼动追踪数据识别用户特征和状态已成为可能。尽管研究兴趣日益浓厚，但一直缺乏对基于眼睛的识别方法的全面系统综述。本研究旨在遵循PRISMA 2020指南，通过系统回顾和综合现有关于使用眼动追踪数据基于机器学习识别用户特征和状态的文献来填补这一空白。纳入标准侧重于那些应用眼动追踪数据通过机器学习或深度学习方法识别用户特征和状态的研究。在ACM数字图书馆和IEEE Xplore中进行了检索，并使用标准方法学标准对所发现的研究进行了偏倚风险评估。数据综合包括一个概念框架，该框架涵盖了任务、背景、技术和数据处理以及识别目标。总共纳入了90项研究，这些研究涵盖了各种任务（如视觉、驾驶、学习）和背景（如计算机屏幕、模拟器、自然环境）。识别目标包括认知和情感状态（如情绪、认知工作量）以及用户特征（如个性、工作记忆）。应用了一系列不同的机器学习技术，如支持向量机（SVM）、随机森林和深度学习模型来识别用户状态和特征。本综述确定了最先进的方法和差距，突出了建立最佳实践、更大规模数据集以及使任务和背景多样化的必要性。未来的研究应侧重于提高生态效度、用于稳健用户建模的多模态方法以及开发注视自适应系统。
