Seungyeon Jwa

Hello 🤗, I am a M.S./Ph.D. student in Artificial Intelligence at Seoul National University and a member of the SNU Machine Perception and Reasoning Lab., advised by Prof. Jonghyun Choi.

I am broadly interested in evaluating and training language models with systematic, data-centric approaches.
My research focuses on:

• Automated, reliable, and scalable evaluation of LLMs
• Data synthesis and curation for training LLMs
• Quantifying and understanding LLM behaviors

Email | CV | Google Scholar | GitHub

Publications

Becoming Experienced Judges: Selective Test-Time Learning for Evaluators
Seungyeon Jwa, Daechul Ahn, Reokyoung Kim, Dongyeop Kang, Jonghyun Choi
EACL 2026 (short)
Paper (long version)

OffsetBias: Leveraging Debiased Data for Tuning Evaluators
Junsoo Park*, Seungyeon Jwa*, Meiying Ren, Daeyoung Kim, Sanghyuk Choi
Findings of EMNLP 2024
Paper | Code | 🤗 Collection | Model | Dataset | Blog

Mitigating Hate Speech in Korean Open-domain Chatbot using CTRL
Seungyeon Jwa*, Young-rok Cha*, Moonsu Han, Donghoon Shin
HCLT 2021
Paper

Work Experience

NCSOFT — NLP Researcher
Application LM Team (Alignment Team) · Feb 2024 – Feb 2025

• Designed a debiasing data construction pipeline to mitigate evaluator bias
• Trained a bias-robust generative judge model, improving the reliability of automatic LLM evaluation

Dialogue Model Team · Jan 2022 – Jan 2024

• Constructed instruction datasets for training dialogue models aligned with user intent and system personas
• Developed an emotional state simulator enabling emotion-aware response generation

NCSOFT — NLP Research Intern
Dialogue Model Team · Jan 2021 – Aug 2021

• Conducted research on controllable dialogue generation
• Proposed a control-token-based method for mitigating hate speech in dialogue models