Date: March 12, 2024

Time: 12:00pm – 1:00pm

Location: Zoom

Abstract:

The idea of a construct is central in the psychological and social sciences: constructs are abstract categories like empathy, misinformation, or benefits of social interaction that are operationalized in order to make them measurable. Social scientists spend an enormous amount of time and care in thinking about constructs; they are the vocabulary over which theories are defined (Bhattacherjee, 2019). This stands in striking contrast to a great deal of current computational research in AI and NLP, where theory tends to be secondary and arguably most work utilizes directly observable behaviors (e.g., did a Twitter user share this URL or not?), simplistic operational definitions (e.g. equating the notion of emotion with labels corresponding to Ekman’s (1999) proposed “basic” emotions; cf. Barrett (2017)), or intuitive but poorly-defined proxy relationships (e.g. including emoji in an analysis, under the assumption that they map transparently to actual emotions). The gap between social science needs and computational practice is deeply problematic: computer scientists often blast forward with highly scalable methods without establishing careful connections to real-world problems and research questions, while social scientists often expend effort on traditional manual analysis out of a distrust for automated solutions or, conversely, they use the output of computational systems uncritically as if what they produce is known to be correct.

This research project’s challenge is to bridge that gap by advancing a recent AI approach called few-shot learning, bringing together cutting-edge machine learning with human expertise in a new way. Paired with that methodological goal is substantive research on two societally high impact problems in social and psychological science. One is to pursue a deeper understanding of disparate impacts of COVID-19 across demographic groups by using survey methods with open-ended text responses; the other is to identify constructs associated with Suicide Crisis Syndrome (SCS; Yaseen et al., 2019) based on people’s language use.

SoDa Seed Grants: The projects under this initiative may address any societal challenge that affects a large number of people, including but not limited to health, public safety, justice, race, gender, education, employment, transit, and political representation.  The goal of these seed grants is to encourage faculty to develop collaborative projects that stimulate the advancement of new ideas that can build the university’s expertise toward a national reputation in the broad area of social data science. The projects blend the development or use of innovative data science methods or new measurements, the advancement of scholarship within or across disciplines, and progress in addressing a societal challenge. 

The SoDa Center at UMD SoDa Symposia highlight the diverse challenges and opportunities in the emerging area of Social Data Science.   Combining insights of SoDa researchers and partners from UMD and around the world, these regular virtual events showcase research and expert commentary about advances and open problems in the use of surveys, administrative, and trace data to understand and shape the social world we live in. Ranging from technical challenges of gathering high-quality data, ethical management of social data at scale, or examples of the power of social data science in education, business, government, or civic life, SoDa Symposia provide an opportunity for a broad audience of researchers, students, and practitioners to learn more about the potential of social data science to change the world.