Doctoral Defense - Michelle Kim

About the Event

The Department of Computer Science & Engineering
Michigan State University
Ph.D. Dissertation Defense

July 10th, 2025 at 9:30 AM EST
Zoom: Information Upon Request from Vincent Mattison or Advisor

Interdisciplinary Decoding of Social Biases Across Domains
By: Michelle Kim
Advisor: Dr. Kristen Johnson

Computational social science integrates computational tools and approaches into the study of social phenomena. This integration of computational methods addresses a critical need within social science research. With the increasing influence of artificial intelligence (AI) and machine learning (ML), computational social science approaches allow researchers to model, simulate, and analyze social phenomena on scales previously unattainable through traditional methods.

While many social phenomena exist, this thesis concentrates on the issue of biases, particularly in the context of Natural Language Processing (NLP). Large language models (LLMs), fundamental to mainstream NLP approaches, perpetuate biases present in the training text corpora. These biases manifest in various forms, including gender, age, sexual orientation, ethnicity, religion, political spectrum, and culture, resulting in discriminatory language generation.

We explore the presence of social biases and stereotypes across various domains of NLP, adopting an interdisciplinary lens that bridges computer science with other fields. By utilizing established theories and developing theory-driven computational algorithms, we seek to enhance our understanding of these complexities.

In this dissertation, we illustrate the advantages of integrating interdisciplinary insights into computational methods. By detecting and analyzing biases, we aim to comprehend the impact of biases in LLMs' implementation and devise methods to mitigate their influence on shaping social perceptions and perpetuating harmful stereotypes.