Title: Computing for Social Science: Characterizing, Quantifying, and Analyzing Social Phenomena in Technology Mediated Communications

Clayton (“C.J.”) Hutto

Human Centered Computing Ph.D. Candidate

School of Interactive Computing

College of Computing

Georgia Institute of Technology

Date: June 20th, 2018 (Wednesday)
Time: 1:00pm to 3:00pm (Eastern Time)
Location: TSRB Conference Room 223

Committee:
---------------

Dr. Eric Gilbert (Advisor, School of Interactive Computing, Georgia Institute of Technology)

Dr. Amy Bruckman (School of Interactive Computing, Georgia Institute of Technology)

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

Dr. Erica Briscoe (Chief Scientist, Advanced Systems Laboratory, Georgia Tech Research Institute)

Dr. Phillip Odom (ML and Analytics, Advanced Systems Laboratory, Georgia Tech Research Institute)

Dr. Scott Counts (Social Technologies Group, Microsoft Research)

Abstract:
-----------

Traditional social science methods rely on labor and time intensive methods to transform qualitative data into quantitative representations of phenomena of interest. In order to rapidly conduct such social scientific research on larger scales, social science researchers need to incorporate computational tools and methods. The Computational Social Science (CSS) paradigm offers useful perspectives for gaining insights from large-scale analyses of demographic, behavioral, social network, and technology-mediated social communication data to investigate human activity, relationships, and other phenomena at multiple scales (e.g., individual, organizational, community, social group, and societal). Human Centered Computing (HCC) complements CSS by offering foundational science for designing, developing, evaluating, and deploying computational artifacts that better support the human endeavors associated with the conduct and practice of CSS research. This dissertation demonstrates theoretical, methodological, and technological contributions resulting from blending traditional social science with computational approaches for the study of human behavior.

Following the CSS paradigm, I build theoretically-informed representations of social constructs—e.g., models of interpersonal relationships and the complex cognitive processes related to human perceptions of sentiment and bias—then use HCC principles to develop and evaluate computational tools that implement those models for the purpose of aiding social science research oriented around large-scale content analysis (e.g., of content from social media networks, product and movie reviews, and newspapers). I do this first by presenting insights from characterizing, quantifying, and analyzing persistent social tie formations in a popular online social network, Twitter. Next, I delve deeper into the analysis of text-based social media content by applying human-centered methods to develop, evaluate, and deploy a computational model (called VADER) to support large scale sentiment analysis of online content from social media, news articles, and user-generated reviews of movies and products. I support the development and evaluation of VADER (and similar CSS- and HCC- inspired technology) by further refinement of a generalized methodological framework for conducting high-volume human evaluations/validation on large scales without jeopardizing qualitative data analysis quality. Finally, I apply the methods, tools, and techniques described above to define and develop a computational model to detect and quantify the degree of perceived bias in journalistic news stories.

In short, this dissertation presents the confluence of computational social science perspectives for theory building and application with human-centered development, evaluation, and deployment of computational tools and methods to support the large scale systematic and (unobtrusive) study of social phenomena as observed via technology-mediated communications and interactions in online content.