SCH: Detecting and mapping stress patterns across space and time: Multimodal modeling of individuals in real-world physical and social work environments

压力已被确定为21世纪的健康流行病，与办公室相关的工作是美国人压力的重要驱动因素，因为长时间工作、快截止日期、工作量大、工作不安全。然而，办公室职员往往完全没有意识到压力的影响，直到他们注意到身体或精神健康或幸福感下降的症状，如肌肉骨骼不适、头痛、睡眠不足或缺乏动力。更有问题的是，大多数人不知道他们的工作活动以及物理和社会工作环境与压力和其他健康结果之间的关系。虽然压力几乎总是被视为不利的，但压力也可以是积极的。我们有机会更好地理解如何促进对生产力有活力和必要的压力，并尽量减少导致负面情绪、不安的身体状态、紧张和倦怠的痛苦。因此，该项目旨在生成新的分析模型，以揭示和绘制影响工作压力的模式和途径，从而了解跨空间和时间压力的主要影响因素。该项目将制定方法，整合来自环境、个人和其他现有技术的不同类型数据，以确定模式，为个性化解决方案提供信息，以提高自我意识和管理与工作有关的健康和福祉。通过对压力和痛苦进行更深入的个性化理解和检测，该项目将影响和促进工作场所的健康和福祉。该项目将作为开发嵌入智能工作场所的传感系统的基础，以自动化环境支持或提供行为反馈。这些影响不仅会改善工人的健康和福祉，而且可以支持减少工人缺勤和提高生产力。因此，该项目有可能改变促进和实现办公室健康和福祉的方式，让员工参与到他们的健康和福祉中来，并最终减少因压力造成的社会和经济损失。这项工作还将对NSF感兴趣的几个标准产生更广泛的影响。它将提高人们对建筑、社会和工作环境对健康和福祉的影响的认识，鼓励K-12学生从事科学和工程方面的职业。它将通过将研究成果纳入跨学科的课程，并通过出版物、演讲和其他媒体传播研究成果，加强研究和教育的基础设施。该项目将使用利益相关者参与、交易的方法来描述个性化的压力体验，并使用广泛的生物行为、环境和活动参与传感技术开发多模式模型，以确定最有价值的数据组合，为员工从事日常工作时的压力管理提供个性化、自动化或技术支持的干预方法。为了建立对办公室工作人员压力的个体情境化理解，机器学习方法可以在多个时间分辨率下处理异构和嘈杂的多模态数据流，包括允许无监督地发现行为例程。个人访谈和生态瞬间评估（EMA）调查将用于描述每个参与者，他们的工作，以及他们如何理解压力的概念（即，痛苦和压力），特别是与他们的工作有关。将对移动和可穿戴技术进行评估，以了解员工在不同时间从事不同工作空间（例如，家庭、正式、公共场所）时的压力体验。可以嵌入正式工作空间的传感方法，以获得替代的、补充的或额外的数据，这些数据对确定工人的压力经验有用，将被评估以区分工人的痛苦和压力。具体来说，物理环境、任务参与、姿势和工人情绪状态对理解压力的贡献将被检查。此外，通过焦点小组，将引出用户的见解、反馈和偏好，这项工作将提高我们对工作环境中健康技术的接受程度，以及它如何与压力/健康自我管理（包括隐私、可信度、接受程度、首选/适当的反馈或自动化方法）相互作用的认识。新的机器学习方法将被开发并用于从多模态数据中预测积极和消极压力，这些数据包括行为特征和基线状态的参考评估，包括与压力、影响和工作相关的参考评估，这些数据作为建模的结构。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Stress has been identified as the health epidemic of the 21st century, and office-related work is a significant driver of stress among Americans due to long hours, rapid deadlines, heavy workload, and job insecurity. Yet, office workers are often entirely unaware of the impact of stress until they notice symptoms of declining physical or mental health or well-being, such as musculoskeletal discomfort, headaches, poor sleep, or lack of motivation. Even more problematic, most individuals do not know how their work activities and the physical and social work environments are related to stress and other health outcomes. While stress is almost always treated as unfavorable, stress can be positive. Opportunities exist to better understand how to promote eustress that is energizing and essential for productivity, and minimize distress that leads to negative emotions, disturbed bodily states, strain, and burnout. Thus, this project aims to generate new analytic models to uncover and map the patterns and pathways that influence work-related stress to understand the primary contributing factors to stress across space and time. The project will develop methods for integrating different types of data from the environment, the person, and other existing technologies to identify patterns that inform personalized solutions for improving self-awareness and management of work-related health and well-being. By developing a deeper individualized understanding and detection of eustress and distress, this project will impact and advance workplace health and wellness. The project will serve as a foundation for the development of sensing systems embedded within smart workplaces to automate environmental supports or provide behavioral feedback. These impacts will not only lead to improved worker health and well-being but can support decreased worker absenteeism and improved productivity. Thus, the project has the potential to change the way health and well-being are promoted and achieved in the office by engaging the worker in their health and wellness and ultimately reducing social and financial losses due to stress. The work will also have broader impacts regarding several criteria of NSF interest. It will promote awareness of the effects of the built, social, and work environments on health and well-being to encourage K-12 students to pursue careers in science and engineering. It will enhance the infrastructure for research and education by incorporating findings into the curriculum across multiple disciplines and disseminating findings via publications, presentations, and other media. The project will use a stakeholder-engaged, transactional approach to describe individualized experiences of stress and develop multimodal models using a wide range of bio-behavioral, environmental, and activity engagement sensing technologies to identify the most valuable combinations of data that inform personalized, automated, or technology-supported intervention approaches to stress management as workers engage in their daily work. To build an individually contextualized understanding of stress among office workers, machine learning methods that can operate with heterogeneous and noisy multimodal data streams at multiple temporal resolutions, including enabling unsupervised discovery of behavioral routines will be developed. Individual interviews and ecological momentary assessment (EMA) surveys will be used to characterize each participant, their work, and how they understand the concepts of stress (i.e., distress and eustress), particularly related to their work. Mobile and wearable technologies will be evaluated to understand stress experiences as workers engage in different workspaces (e.g., home, formal, public) across time. Sensing methods that could be embedded within the formal workspace to obtain alternative, complementary, or additional data useful in determining experiences of worker stress will be evaluated for differentiating worker distress from eustress. Specifically, the contribution of the physical environment, task engagement, posture, and worker emotive states to the understanding of stress will be examined. Additionally, through focus groups that will elicit user insights, feedback, and preferences, the work will advance our knowledge about acceptance of technology for health in work settings, and how that interacts with stress/health self-management including privacy, trustworthiness, acceptance, preferred/appropriate methods for feedback or automation. Novel machine learning methods will be developed and employed to predict positive and negative stress from multimodal data that include reference assessments of behavioral traits and baseline states–including those related to stress, affect, and the job–that serve as constructs for modeling.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.