Title: Computational and Causal Approaches on Social Media and Multimodal Sensing Data: Examining Wellbeing in Situated Contexts

Koustuv Saha

Ph.D. Candidate in Computer Science

School of Interactive Computing

Georgia Institute of Technology

https://koustuv.com/

Date: Thursday, June 03, 2021

Time: 3:30 PM - 6:00 PM ET

URL: https://bluejeans.com/514937767

Committee:

--------------

Dr. Munmun De Choudhury (Advisor, School of Interactive Computing, Georgia Institute of Technology)

Dr. Gregory D. Abowd (School of Interactive Computing, Georgia Institute of Technology | College of Engineering, Northeastern University)

Dr. Thomas Plötz (School of Interactive Computing, Georgia Institute of Technology)

Dr. Emre Kıcıman (Information and Data Sciences, Microsoft Research)

Dr. Gloria Mark (Donald Bren School of Information and Computer Sciences, University of California, Irvine)

Abstract:

------------

 A core aspect of our lives is often embedded in the communities we are situated in. The interconnectedness of our interactions and experiences intertwines our situated context with our wellbeing. A better understanding of wellbeing will help us devise proactive and tailored support strategies. However, existing methodologies to assess wellbeing suffer from limitations of scale and timeliness. These limitations are surmountable by social and ubiquitous technologies. Given its ubiquity and wide use, social media can be considered a “passive sensor” that can act as a complementary source of unobtrusive, real-time, and naturalistic data to infer wellbeing. This dissertation leverages social media in concert with multimodal sensing data, which facilitate analyzing dense and longitudinal behavior at scale. In this dissertation, I adopt machine learning, natural language, and causal inference analysis to infer the wellbeing of individuals and collectives, particularly in situated communities, such as college campuses and workplaces.

Before incorporating sensing modalities in practice, we need to account for confounds. One such confound that might impact behavior change is the phenomenon of “observer effect” --- that individuals may deviate from their typical or otherwise normal behavior because of the awareness of being “monitored”. I study this problem by leveraging the potential of longitudinal and historical behavioral data through social media. Focused on a multimodal sensing study, I conduct a causal study to measure observer effect in social media behavior and explain the observations through existing theory in psychology and social science. The findings provide recommendations to correcting biases due to observer effect in social media sensing for human behavior and wellbeing.

The novelties and contributions of this dissertation are four-fold. First, I use social media data that uniquely captures the behavior of situated communities. Second, I adopt theory-driven computational and causal methods to make conclusive research claims on wellbeing dynamics. Third, I address major challenges with methods to combine social media with multimodal sensing data for a comprehensive understanding of human behavior. Fourth, I draw interpretations and explanations of online-data-driven offline inferences. This dissertation situates the findings in an interdisciplinary context, including psychology and social science, and bears implications from theoretical, practical, design, methodological, and ethical perspectives catering to various stakeholders, including researchers, practitioners, and policymakers.