Human Behavior in Context: Analyzing Routines and Interactions Through Multi-Source Behavioral Data

The availability of new data sources following the digital revolution has greatly enhanced our ability to study human behaviour in real-world contexts. This development has given rise to Computational Social Science, a new discipline that aims to uncover patterns in human behaviour through an inherently multidisciplinary approach. From analyzing collective behaviour with complex networks, to developing computational models for social science questions, researchers have come together to provide a more comprehensive understanding of how we act as individuals and as communities. In this context, we investigate human behaviour using multiple data sources with varying level of contextual and environmental information. On one side, we study it through the lens of Living Labs, experiments that use sensors and mobile devices to collect behavioural data from users in real or realistic environments. Here, we focus first on the social dimension of human behaviour by modelling face-to-face interactions and examining how individuals organize into communities. Specifically, we propose a modelling approach to enrich these datasets, which often lack temporal depth or have a limited number of participants due to the high costs of data collection. We then analyze students’ daily schedules using a dataset richer in context, which includes activity and location information along with details about with whom they interacted, thanks to self-reported Time Diaries. We show how this richer context is important for describing their behavioural shifts in response to the COVID-19 pandemic. On the other side, we examine human behaviour through the lens of mobility data. With the wide availability of GPS data, the study of how people commute and live in cities has been transformed, from analysing mobility patterns and distances travelled, to transportation usage, to detecting colocations that indicate social gatherings. Visits to Points of Interest (POIs)} can also be used to construct sequences of daily activities, such as going to a bar or seeing a movie, similarly to the aforementioned time use diaries. In this way,we can address several challenges inherent to data collection in living labs, such as limited participant numbers, constrained spatial environments, and short observation windows. In this work, we combine data from the American Time Use Survey (ATUS), which provides time use diaries for thousands of individuals, with GPS data provided by a location intelligence company. By applying insights into human routines, we investigate the biases that may arise when using mobility data to infer daily activities, and we highlight the need for caution when employing such data in contemporary research.