Networks are widely used to understand complex real-world systems. This thesis focuses on social networks in which nodes are individuals and edges are individuals’ friendship, family, or professional relationships. Social networks play an important role in individuals’ physical and mental health. Traditional research aiming to link individuals’ social networks and health has focused on studying a relationship between static network structure and static health-related traits due to the inability to collect individuals’ dynamic social interaction data and dynamic health-related trait data. With recent advancements of data collection technologies, such as smartphones and wearable sensors, one is able to collect such data. This is exactly what we have done in the NetHealth study, which gathered social interaction data from smartphones (i.e., SMS communications), health-related behavioral data from wearable sensors (i.e., Fitbit data), and individuals’ trait data from surveys (e.g., personality traits and mental health) of around 700 Notre Dame undergraduates during 2015 to 2019.
Leveraging the rich NetHealth data, this thesis focuses on uncovering relationships between individuals’ social network positions, health-related behaviors, and various other traits and developing network-based models to predict individuals’ mental health. In particular, we look at the co-evolution of individuals’ social network positions (i.e., centralities) and their behaviors (i.e., physical activities), with the goal of studying whether groups of individuals who have similar evolving social network profiles or similar evolving physical activity profiles (or both) share similar traits such as personality, depression, and anxiety. We are the first ones to study the relationship between individuals’ positions in a dynamic social network and their dynamic health-related behaviors. Our results reveal several associations between individuals’ social network structure, health-related (i.e., physical activity) behaviors, and other (e.g., personality or mental health) traits. So, in a follow-up study, we integrate the different data types from the NetHealth study into a heterogeneous information network (HIN) to develop a predictive model of social network structure from behavioral/trait information or vice versa. Specifically, we focus on the task of predicting one’s mental health (i.e., likelihood of being depressed or anxious) from the rest of the data. In this context, we are the first ones to define the problem of predicting individuals’ mental health as applying to our HIN a popular paradigm of a recommender system (RS), which is typically used to predict the preference that an individual would give to an item (e.g., a movie or book). In our case, the items are the individuals’ different mental health states. Since the existing RS methods work on static network data and extending them to dynamic network data is non-trivial, our constructed HIN aggregates the dynamic social network data from the considered study time period into a static network. And while we show that even this leads to high mental health prediction accuracy, at least some temporal information is lost. Including as much of temporal information as possible could further improve prediction accuracy. With this hypothesis, in a follow-up study, we fairly evaluate whether using dynamic social network data has more power than using static social network data in the task of predicting mental health. Indeed, we find this to be the case. We are the first ones to develop predictive models of mental health that use dynamic social network data.