Algorithms for Quantifying Human Behavioral Dynamics from Smartphone Sensor Data

计算机和计算机系统并不独立于人类行为。我们所处的位置和我们所做的工作推动了物理和虚拟基础设施中的布局和算法。这些行为的影响从计算机科学延伸到土木工程和建筑(我们在哪里，我们如何在太空中移动？)，公共卫生领域(我们接触了多长时间，与谁接触？)，甚至社会科学(空间如何调节流动性？)。这些行为中的许多都植根于我们所处的位置，我们与谁在一起，以及我们正在做什么。了解人们在哪里以及他们是谁有助于建模人员更好地了解空间的使用或疾病的传播，无线网络设计人员希望提供更好、更具响应性的服务，甚至社交机器人也知道他们可能会在哪里以及何时会见特定的个人。 直到最近，收集这些数据一直受到技术的阻碍。社会科学家的传统工具要么过于生硬(例如调查或人口普查数据)，要么过于昂贵而无法广泛使用(例如直接观察或人种学)。传感器微型化和智能手机集成的最新进展，使得收集关于人类行为的地点、时间、方式和与谁的高保真数据成为自动化。在过去的四年里，我们的实验室在设计和部署这类系统方面取得了相当大的成功。 然而，收集数据只是冰山一角。千兆字节的行为数据需要以这样一种方式进行处理，以便研究人员能够以导致洞察力的方式操纵数据。这项拨款建议研究我们如何以合理的方式对人类的行为进行编码。就像湍流的一般行为可以用一个或两个简单的比率来概括一样，我们希望推导出人类行为的指标，并为计算这些指标提供高效的计算算法。 这项研究将产生深远的影响，改变有关公共卫生和基础设施的决策方式，并提供一种新的方式来理解作为人类的样子。

Computers and computer systems are not independent of human behaviour. Where we are and what we do drives placement and algorithms in both physical and virtual infrastructure. The impact of these behaviours extend beyond computer science to civil engineering and architecture (where we are and how we move through space?), public health fields (for how long and with whom have we been in contact?), and even the social sciences (how is mobility mediated by space?). Many of these behaviours are anchored in where we are, who we are with, and what we are doing. Understanding where people are and who they are with helps modelers who seek to better understand the use of space or the spread of disease, wireless network designers looking to provide better more responsive service, and even social robots understand where and when they might meet specific individuals. Until recently, gathering this data has been hampered by technology. The traditional tools of the social scientist are either too blunt (e.g. surveys or census data) or too expensive to deploy widely (e.g. direct observation or ethnographies). Recent advances in the miniaturization of sensors, and their incorporation into smartphones, has made collecting high fidelity data on the where, when, how and with whom of human behaviour automatable. Our lab has had considerable success over the last four years designing and deploying these kinds of systems. However, collecting data is just the tip of the iceberg. Gigabytes of behavioural data need to be processed in such a way as to allow researchers to manipulate the data in ways that lead to insight. This grant proposes to investigate how we might encode the behaviour of humans in reasonable ways. Just as the general behaviour of a turbulent water flow can be encapsulated in one or two simple ratios, we hope to derive metrics of human behaviour, and efficient computational algorithms for computing them. This research will have profound and far-reaching impacts, changing the way that decisions about public health and infrastructure are made, and providing a new way of understanding what it looks like to be human.