Research

Published Papers

• A Benchmark Dataset for Evaluating Gender Sensitivity in Korean Political Discourse with LLMs
  Scientific Data 2025
  With Sanghoun Song, Sunkyoung Park, Eunbi Cho, Chanyoung Jung, Woochang Kang, Eunah Park
  
  Abstract

  Large language models (LLMs) are increasingly used to analyze political discourse, yet their ability to detect subtle, culturally grounded variations in gender sensitivity remains underexplored. We introduce KOGENT, a benchmark dataset of 1,222 transcripts from the Korean National Assembly, annotated for gender sensitivity across 6,024 utterances. Each utterance is labeled as high or low in gender sensitivity, based on contextual indicators of bias, discrimination, or inclusion, and tagged for the target group (e.g., women, men, sexual minorities, all genders). KOGENT covers legislative sessions from 1948 to 2024, including plenary sessions, committees, and hearings. Annotation reliability was ensured through dual coding and adjudication, yielding high intercoder agreement (Cohen’s κ = 0.96; Krippendorff’s α = 0.95). When tasked with labeling utterances by gender sensitivity, GPT-4.1 achieved F1-scores of 87.5% (zero-shot) and 91.2% (18-shot) for high gender sensitivity labels, while GPT-4o reached 90.4% and 91.1%, respectively. While incorporating in-domain examples enhanced model performance, limitations in distinguishing between criticisms and reinforcements of inequality, culturally specific terminology, and extended contexts were observed for both models. Our results demonstrate KOGENT’s utility as a robust benchmark for analyzing gender sensitivity in Korean political speech and evaluating multilingual LLMs’ sociocultural alignment.
• Legacies of Party Origins on the Electoral Performance of Former Authoritarian Ruling Parties
  Democratization 2025
  With Wonjun Song
  
  Abstract

  In this paper, we examine why some former authoritarian ruling parties decline after democratic transitions while others survive and thrive. We argue that former ruling parties that develop strong party brands prior to the onset of authoritarian rule (pre-seizure) are better positioned to garner electoral support under democracy than those created after dictators consolidated power (post-seizure). Pre-seizure parties, by building a party brand before authoritarian onset, are more capable of compensating for the loss of state resource advantages under democracy, whereas post-seizure parties reliant on material inducements are more vulnerable to collapse. Leveraging a dataset covering former authoritarian ruling parties from 1946 to 2010, we find that pre-seizure parties are more likely to avoid dissolution and achieve higher legislative vote shares after the democratic transition. They also exhibit greater resilience to transitional economic crises, as the negative effects of such crises on electoral performance are significantly mitigated. Complementary analysis also shows that pre-seizure parties rely more on grassroots support and less on clientelism, consistent with characteristics associated with stronger party brands. Together, these findings underscore the lasting impact of party origins on post-transition trajectories and offer new insights for the study of comparative authoritarianism, democratization, and political party development.
• Is External Information Useful for Stance Detection with LLMs?
  Findings of the Association for Computational Linguistics 2025
  With Quang Minh Nguyen
  
  Abstract

  In the stance detection task, a text is classified as either favorable, opposing, or neutral towards a target. Prior work suggests the use of external information, e.g., excerpts from Wikipedia, improves stance detection performance. However, whether or not such information can benefit large language models (LLMs) remains an unanswered question, despite their wide adoption in many reasoning tasks. In this study, we conduct a systematic evaluation on how external information can affect stance detection across eight LLMs and in three datasets with 12 targets. Surprisingly, we find that such information degrades performance in most cases, with macro F1 scores dropping by up to 15.9%. This degradation is even more pronounced at a 28.1% drop when stance biases are introduced in the external information, as LLMs tend to align their predictions with the stance of the provided information rather than the ground truth stance of the given text. We also find that fine-tuning mitigates bias but does not fully eliminate it. Our findings, in contrast to previous literature on BERT-based systems suggesting that external information enhances performance, highlight the risks of information biases in LLM-based stance classifiers.
• The Effects of Partisan Elites’ Violent Rhetoric on Support for Political Violence
  Political Behavior 2025
  Abstract

  Violent partisan hostility is undermining American democracy. How does partisan elites’ violent rhetoric shape support for political violence? Focused on social media communication where individuals are exposed to elite messages from both sides of the partisan divide, I conduct an online experiment to examine the impact of co-party and opposing party elites’ violent rhetoric on support for political violence and the medating role of emotions in the process. Drawing insights from theories of opinion leadership and inter-group conflict, I demonstrate that co-party (but not opposing party) elites’ violent rhetoric increases support for violence and that partisans fail to countervail against elites’ violent rhetoric. Further, I show that fear mediates the inflaming effect whereas anger, disgust, and sadness suppress it. This paper is among the first to make sense of the effects of elite rhetoric on violent partisan hostility, advancing knowledge in political violence, political communication, and political psychology.
• The National Network of U.S. State Legislators on Twitter
  Political Science Research and Methods 2024
  With Ishita Gopal, Nitheesha Nakka, Jeffrey Harden, Frederick Boehmke, Bruce Desmarais
  
  Abstract

  A lot of attention has been paid to studying the online activity of the members of the United States Congress. This scrutiny has not been extended to state legislators. Very few studies exist which catalogue why state legislators connect and communicate with one another online in the ways they do. Inspired by this question and building on studies which have analyzed online communication of members of national legislatures, this paper aims to systematically analyze state legislator relationships in the online environment. We collect original data for 4000+ legislators and study patterns of connection and communication of state legislators on Twitter. The results from this study will help better understand what motivates tie formation in the online environment and if these patterns of connection conform to or can predict offline relationships. We test the impact of variables such as party affiliation, state, chamber, cohort, gender, and policy area focus on the organization of these online networks. We look at three main types of networks that can arise due to participation on Twitter - follower, retweets and mentions. We also aggregate the ties to infer dynamics between states.
• Violent Political Rhetoric on Twitter
  Political Science Research and Methods 2023
  * Recipient of 2021 John Sprague Award from the Political Networks Section, American Political Science Association
  Abstract

  Violent hostility between ordinary partisans is undermining American democracy. Social media is blamed for rhetoric threatening violence against political opponents and implicated in offline political violence. Focusing on Twitter, I propose a method to identify such rhetoric and investigate substantive patterns associated with it. Using a data set surrounding the 2020 Presidential Election, I demonstrate that violent tweets closely track contentious politics offline, peaking in the days preceding the Capitol Riot. Women and Republican politicians are targeted with such tweets more frequently than non-Republican and men politicians. Violent tweets, while rare, spread widely through communication networks, reaching those without direct ties to violent users on the fringe of the networks. This paper is the first to make sense of violent partisan hostility expressed online, contributing to the fields of partisanship, contentious politics, and political communication.
• Attention to the COVID-19 Pandemic on Twitter: Partisan Differences among U.S. State Legislators
  Legislative Studies Quarterly 2022
  With Nitheesha Nakka, Ishita Gopal, Bruce Desmarais, Abigail Mancinelli, Jeffrey Harden, Hyein Ko, Frederick Boehmke
  
  Abstract

  Subnational governments in the United States have taken the lead on many aspects of the response to the COVID-19 pandemic. Variation in government activity across states offers the opportunity to analyze responses in comparable settings. We study a common and informative activity among state officials—state legislators’ attention to the pandemic on Twitter. We find that legislators’ attention to the pandemic strongly correlates with the number of cases in the legislator’s state, the national count of new deaths, and the number of pandemic-related public policies passed within the legislator’s state. Furthermore, we find that the degree of responsiveness to pandemic indicators differs significantly across political parties, with Republicans exhibiting weaker responses, on average. Lastly, we find significant differences in the content of tweets about the pandemic by Democratic and Republican legislators, with Democrats focused on health indicators and impacts, and Republicans focused on business impacts and opening the economy.

Selected Working Papers

• Public Policymakers’ Online Anti-vaccine Statements: The Important Role of Engagement
  Revise and Resubmit
  With Xinyu Wang, Sarah Rajtmajer, Jeffrey Harden, Frederick Boehmke, Bruce Desmarais
  

• Networked Policy Diffusion in the U.S. States: The Case of COVID-19 Response
  Revise and Resubmit
  With Frederick Boehmke, Bruce Desmarais, Ishita Gopal, Jeffrey Harden, Samuel Harper, Johabed Olvera, Issac Polbert
  

• Rhetoric for the Win: The Strategic Use of Morality on Social Media
  Under Review
  With Juyeon Julia Park
  

• Uneven Automation: AI’s Impact on Software Tasks Varies by Difficulty and Data Availability
  Under Review
  With Myokyung Han, Jiwoon Hong, Taegyoon Kim, Jinhyuk Yun, Lanu Kim
  

• Difficulty Perception in the Reasoning of LLMs
  Under Review
  With Quang Minh Nguyen
  

• From Empathy to Risk Management: Support Seeking in Human-LLM Interaction
  Under Review
  With Joohee Choi, Jeongwoo Jang
  

• Fighting for or Against Freedom? The Geopolitical Paradox of South Korea’s Impeached President
  Under Review
  With Jiyoung Baek, Seungook Lee
  

• Diffusion in a Time of Uncertainty: How the Determinants of State Policy Choices Changed During the Pandemic
  Under Review
  With Frederick Boehmke, Bruce Desmarais, Jeffrey Harden, Samuel Harper, Hyein Ko, Johabed Olvera, Issac Polbert, Tracee Saunders, Taylor Tokos
  

• How Does Toxicity Emerge in Political Livestreams? Large-Scale Evidence from Twitch
  Under Review
  With Hayden Arnold, Sangyeon Kim
  

• Polarization and Science in Public Policymaking: An Analysis of U.S. Policy Documents
  With Alexander Furnas, Dashun Wang
  

• Evidence Outperforms Intuition in Changing Minds Online: Insights from 1.6 Million Online Debates
  With Joohee Choi, Hayden Arnold, Jeongwoo Jang