Exploring the Diversity of Opinions on Affirmative Action Through Extended Stance Detection Among Reddit Users
Abstract: Affirmative Action (AA) has remained a controversial topic in the U.S. for several decades. While previous research has extensively explored AA from legal, social, and ethical aspects, there is a lack of computational work to investigate this topic from the lens of users on social media. By collecting over 2,500 posts from 23 prominent Reddit communities, our study attempts to gain a better understanding of how online users view AA. We build upon the previous work on stance detection, by introducing a new set of stance categories that can efficiently reflect a diverse range of opinions on AA. Finally, we explore the performance of three LLM-based classifiers, powered with four carefully designed prompts to classify these categories and run several analyses to enhance our understanding of the AA discourse. Our findings show that GPT-4 with instruction and examples is the most efficient for classifying stance. We found opposition towards AA more common than support in our dataset with discussions on college admission, and race more prevalent. This study enhances the field by presenting a novel dataset of stances derived from Reddit data and initiates a conversation on broadening binary evaluations of viewpoints on controversial subjects.
Paper Type: Long
Research Area: Computational Social Science and Cultural Analytics
Research Area Keywords: Stance detection- NLP tools for social analysis
Contribution Types: Model analysis & interpretability, Data resources, Data analysis
Languages Studied: English
Submission Number: 4349
Loading