Publications

ASH: A Systematic Automated Evaluation Framework for Creative Text Generation in Cuisine Transfer Tasks

Hoonick Lee, Mogan Gim, Donghyeon Park, Donghee Choi, Jaewoo Kang

IEEE Access (Early Access) · doi:10.1109/ACCESS.2026.3727977

Extends our ICDMW 2025 workshop paper into a full evaluation framework: 4,800 recipes generated across 40 cultures and scored on Authenticity, Sensitivity, and Harmony, with a meta-evaluation of LLM-as-a-judge prompting strategies against human raters from nine countries — where simple scoring rubrics aligned with human judgment better than elaborate chain-of-thought instructions.

An LLM-based Chain-of-Response Counter-Scam System

Heedou Kim, Mogan Gim, Donghee Choi, Hoonick Lee, Soonil Bae, Mi-Young Kim, Jaewoo Kang

International Joint Conference on Artificial Intelligence (IJCAI)

An LLM-based chain-of-response system for detecting online scams and generating safe counter-responses.

Prompting Matters: Snippet-Aware Strategies for Biomedical QA with LLMs in BioASQ 13b

Hajung Kim*, Hoonick Lee*, Yewon Cho*, Jungwoo Park*, Jueon Park*, Soyon Park*, Yan Ting Chok*, Seungheun Baek*, Donghyeon Lee*, Jaewoo Kang

* Equal contribution

CLEF 2025 Working Notes, CEUR Workshop Proceedings, Vol. 4038, pp. 373–382

★ 1st place overall, BioASQ 13b Phase B (Exact Answers)

Snippet-aware prompting strategies for biomedical question answering; the system behind KU-DMIS's first-place overall result in BioASQ 13b Phase B.

Culinary Class Wars: Evaluating LLMs Using ASH in Cuisine Transfer Task

Hoonick Lee, Mogan Gim, Donghee Choi, Jaewoo Kang

2025 IEEE International Conference on Data Mining Workshops (ICDMW), CoGamy Workshop, pp. 657–660

Introduces the ASH (Authenticity, Sensitivity, Harmony) framework to evaluate how LLMs handle culturally grounded creativity in cuisine transfer.

Substituting Radiology Reports for Chest Radiographs does not preserve prediction behavior in Post-Discharge Mortality Prediction

Chanhwi Kim, WonJin Yoon, Hoonick Lee, Jung-Oh Lee, Majid Afshar, Jaewoo Kang, Timothy Miller

JAMIA Open

Compares chest radiographs against their paired radiology reports as inputs for 30-day post-discharge mortality prediction on linked MIMIC-IV/MIMIC-CXR data (n = 1,360). Substituting reports for the images changes how patients are risk-ranked — measured by Kendall's τ distance between predicted rankings — rather than merely lowering discrimination, so expert-written summaries are imperfect prognostic proxies for the images themselves.

Impact of High Pressure and Emotional Prompting Strategies on the Behavior of Large Language Models

Minju Song, Taewhoo Lee, Hyeon Hwang, Yein Park, Chanwoong Yoon, Hoonick Lee, Dain Kim, Sihyeon Park, Jiwoo Lee, Jaewoo Kang

Proceedings of the Korea Computer Congress (KCC), KIISE, pp. 910–912

Studies how high-pressure and emotionally charged prompts change LLM behavior and output quality.

KU-DMIS at EHRSQL 2024: Generating SQL Query via Question Templatization in EHR

Hajung Kim, Chanhwi Kim, Hoonick Lee, Kyochul Jang, Jiwoo Lee, Kyungjae Lee, Gangwoo Kim, Jaewoo Kang

Proceedings of the 6th Clinical Natural Language Processing Workshop (NAACL 2024), pp. 672–686

A question-templatization approach for reliable natural-language-to-SQL generation over electronic health records, built for the EHRSQL 2024 shared task.