October 2026 #Jiyun Beak, #Myokyung Han, and #Lanu Kim published Evaluating Bias in LLM-Synthetic Data across Majority and Minority Groups on General and Stigmatized Topics: The Case of Multicultural and Monocultural Adolescents in South Korea at the 5th Workshop on NLP for Positive Impact, held with EMNLP 2026 in Budapest. Using a survey that gave identical questionnaires to adolescents from multicultural and monocultural families in South Korea, the paper prompts two LLMs with personas matching each respondent to answer 77 survey questions, and finds only marginal differences in replication accuracy between minority and majority groups. Performance varies instead with whether prompts contain socially stigmatized words and with which model is used, highlighting both the potential and the limits of LLM-generated synthetic survey data.
October 2026
Congratulations to #Hyeongjae Lee, Seonwoo Ahn, Sunwoo Kwak, June Jeon, Jisung Yoon, and Taegyoon Kim! They presented Who Counts as “the People”? Measuring Contested Political Concepts from Political Speech using Masked Language Models at EMNLP 2026 in Budapest, where the paper received the SAC Highlight Award. Treating a contested political concept as a combination of theory-based frames (Nation, Civic, and Global), the paper uses a fine-tuned political masked language model to score how U.S. presidential speeches from 1789 to 2025 use the term “the people.” The measurements agree with expert human coding about as closely as human coders agree with each other, trace historical periods such as the Civil War and the Civil Rights era, and relate to external indicators: stronger democracies show less national and more civic framing.
October 2026 #Kyungjong Kim and #Lanu Kim published a paper in The Journal of Police Policies (치안정책연구): A Call Log-Based Early Screening Model for Voice Phishing Risk Numbers: Focusing on Police–Telecommunications Cooperation. Testing detection across cumulative windows of 1 to 30 days of call logs, the paper finds that models trained for a specific window outperform one trained on the full period, that a 7–14 day window offers the most workable balance of speed and reliability, and that behavioral features such as outgoing call ratio and short-call ratio can flag risky numbers on their own. The authors propose using the model as decision support in a police–telecom workflow rather than as automatic blocking.
September 2026 #Myokyung Han, Taegyoon Kim, Jinhyuk Yun, and #Lanu Kim posted a new preprint: The Uneven Decline of Collective Knowledge Production: Evidence from Stack Overflow After Generative AI. Analyzing over two million Stack Overflow questions from 2020-2025, the paper finds that easy questions declined sharply after ChatGPT's release while difficult questions became more common, with the same pattern holding across multiple programming languages: generative AI erodes accessible, straightforward knowledge sharing while preserving complex, specialized content.
September 2026
Congratulations to #Myokyung Han and #Hyeongjae Lee, who were each awarded a 2-year PhD fellowship from the National Research Foundation of Korea (2026.9-2028.8) for their proposals, 'Reconfiguring the Scope of Knowledge Formation through Changing Interaction Patterns in the Generative AI Era: A Focus on Online Knowledge-Formation Communities' (Myokyung Han) and 'The Labor Market Wage Premium Effect of Interdisciplinarity in Higher Education: A Study Using Syllabus Text Data and LLM Analysis' (Hyeongjae Lee)!
August 2026 #Hyeongjae Lee, #Jihyang Cheon, and #Lanu Kim posted a new preprint: Who Delegates to AI? Evidence from 53,000 Agent Configurations. Using a new Agentic Adoption Index built from 53,000 AI agent configurations, the paper finds that real-world AI delegation concentrates in mid-wage, bachelor's-level occupations rather than the jobs earlier exposure models flagged as most at risk, and that adoption still lags behind what is technically feasible.
June 2026 #Kyungjong Kim and #Lanu Kim published a paper in Scientific Reports: Proactive detection of voice phishing networks using call log analysis and machine learning. Using large-scale call log data from South Korea, the paper shows that voice phishing phone numbers have distinctive behavioral signatures, including concentrated weekday activity, shorter calls, more outgoing calls, and heavier use of MVNO lines, letting machine learning models flag them with over 95 percent accuracy.
May 2026
Congratulations to #Maida Aizaz, who successfully defended her master's thesis, 'From Affiliation to Attribution: Who Mentions Whom in Science?' The paper has also been successfully submitted to a journal and is currently under review.
February 2026
Minyoung Choi, #Myokyung Han, #Suhyoung Choi, Eunhee Bae, Dongju Kim, Bong Gwan Jun, and #Lanu Kim published a paper in Asian Journal of Business Ethics: Greedy Leadership Roles: Why Modern Organizations Struggle to Retain Women Leaders. Based on focus groups with women leaders at a South Korean engineering firm, the paper argues that leadership roles themselves, with their demand for near-constant availability, remain a structural barrier to gender diversity even as organizations reduce more overt forms of discrimination.
June 2025
June Jeon, #Lanu Kim, and Jaehyuk Park published a paper in Technology in Society: The Ethics of Generative AI in Social Science Research: A Qualitative Approach for Institutionally Grounded AI Research Ethics. Based on 17 interviews, the paper identifies three narratives researchers use to make sense of generative AI and research ethics (equalizer, meritocracy, and community), and argues that AI research ethics cannot be reduced to universal checklists: ethical dilemmas emerge not from AI's technical functions alone but from how those functions are situated within academia's institutional arrangements, calling for community engagement, education, and institutional governance alongside any technical rules.
November 2024 #Yeongtaek Lim represented South Korea in the Singapore AI Safety Red Teaming Challenge 2024 and achieved third place in the team category through red teaming large language models.
November 2024 #Suhyoung Choi presented research titled 'Exploring When and How Much Annotator Diversity is Necessary for Effective Machine Learning Models' at IC2S2 2024.
November 2024 #Youjin Hong presented research titled 'Contrasting Criminal Judgments: A Comparative Analysis of Publicly Enraging and Indifferent Crimes' at the AKSA Roundtable 2024.