Modeling emotion unobtrusively in interactive digital systems

虽然办公环境是电脑使用的源头，但最近互动系统已经融入我们的家庭、休闲活动和人际关系中。在这些活动中，用户体验在很大程度上受用户情感体验的支配；然而，目前的计算机系统仍然不能理解或适应用户的情绪状态。尽管在情感建模方面取得了进展，但目前的方法（例如，为用户配备生理传感器）通常过于突兀和昂贵，无法在实验室之外使用，因此对许多情感计算应用缺乏实用性。本研究提出的主要目标不是仪器化用户，而是发明对用户不引人注目的情感检测和建模的新方法，从而改进自然环境中的情感检测。我们将通过挖掘用户与其设备和环境交互时留下的现有痕迹来实现这一目标。首先，我们将使用来自计算机交互的痕迹（击键动力学和光标运动学），其次，我们将使用生物特征痕迹（瞳孔扩张，姿势传感），第三，我们将使用来自社交媒体更新和文本消息的语言痕迹。我们还将通过综合多个项目来改进单个模型的性能。这三个主题代表了基础研究，并将推动人机交互、情感计算和情感建模领域的发展。我们将通过应用于三个及时且相关的领域来评估这些模型的有效性：电脑游戏、电脑媒介沟通和幸福感。我们的研究结果将为加拿大人在评估电脑游戏玩家体验到远程诊断新妈妈产后抑郁症等广泛领域提供优势。我们的工作可以改善加拿大人的生活。例如，特定的情感状态可以促进创造力，提高工作满意度，甚至增强我们的免疫反应。使用模型来了解不同触发因素下情绪的变化，可以为改善人们情绪健康的政策和选择提供信息。

Although office environments are the provenance of computer usage, interactive systems have recently woven themselves into our homes, our leisure activities, and our relationships. In these activities, user experience is governed in large part by the user's affective experience; however, current computer systems still do not understand or adapt to the emotional state of the user. Although progress has been made on modeling emotion, current approaches (e.g., instrumenting a user with physiological sensors) are often too obtrusive and expensive for use outside of laboratories, and thus lack utility for many affective computing applications. Rather than instrument a user, the primary objective proposed by this research is to invent novel methods for detecting and modeling emotion that are unobtrusive to the user, thus improving emotion detection in natural settings. We will accomplish this by mining existing traces left behind as users interact with their devices and their environment. First, we will use traces from computer interaction (keystroke dynamics and cursor kinematics), second we will use biometric traces (pupil dilation, posture sensing), and third, we will use linguistic traces from social media updates and text messages. We will also improve individual model performance through synthesis of the multiple projects. These three themes represent foundational research and will advance the fields of human-computer interaction, affective computing, and emotion modeling. We will evaluate the efficacy of the models through application to three timely and relevant domains: computer games, computer-mediated communication, and well-being. Our results will provide advantages to Canadians in areas as broad as evaluating player experience during computer game play to remotely diagnosing post-partum depression in new mothers. Our work can improve the lives of Canadians. For example, specific affective states can promote creativity, improve job satisfaction, and even strengthen our immune response. Using models to understand how emotion changes with different triggers can inform policies and choices that will improve the emotional well-being of people.