PhD Thesis Proposal: Siyan Zhao, "Behavioral Pathways Linking Social Interactions to Mental Health"

Jason I. Hong (CMU)
Robert E. Kraut (CMU)
Mayank Goel (CMU)
Eden Litt (Facebook)

 

Humans are social creatures in nature. We interact with each other and form relationships. On a societal level, these human relations are the basis for social structure. On an individual-level, building and maintaining such close relationships promote stronger feelings of support and connectedness and better well-being. However, existing work on how social relationships impact social connectedness (social support) and well-being is unclear. This is because the social behaviors that mediate the effect of social support and well-being are daily and mostly invisible supportive exchanges. Capturing these exchanges is a challenge. The most common methodology to study social interactions is participants' self-report. However, self-reports are impractical to use throughout a day to log all interactions due to inconvenience. Research has also shown that self-reports of past events can be unreliable and biased, influenced by their mood at the moment of reports. In addition, people are much less likely to report interactions that they consider insignificant or mundane, the exact interactions that researchers would like to capture. For example, a short exchange between partners can be easily overlooked by participants. But this behavior, though seemingly insignificant, can be an invisible source of support as it demonstrates care and willingness to invest in their relationship.

These constraints in method have limited the types of social interactions that researchers can capture, which hinders our attempt to understand the pathway between social connectedness and well-being. In this proposal, I will address these shortcomings using passively collected sensor data on mobile phones. This is an ideal data collection tool as 79% of people between the age of 18-44 have their smartphones with them 22 hours a day. These smartphones, equipped with sensors, have access to user's activities throughout a day. Leveraging this large quantity of data, I can train machine learning models to measure one’s social activities at a finer granularity than self-reports. This richness in data gives me an opportunity to understand which social interactions lead to changes in one’s social connectedness and well-being.

In work done so far, I have identified, from literature, a taxonomy of social interactions characterizing interactions that may have an impact on people. In a longitudinal study, I also demonstrated that phone-mediated interactions, passively collected from smartphones, can predict changes in relationship quality.

For my proposed work, I will focus on building machine learning models that use data collected on mobile phones to measure one's social interactions and their attributes. The final work will shed light on which of the interaction attributes, measured by mobile phone sensors, predict changes in one's social connectedness and well-being.