Language Models
Reasoning, dialogue, memory, personalization, and robust evaluation.
Jinyoung Yeo
Associate Professor
Department of Artificial Intelligence
College of Computing
Yonsei University
Office: 신촌캠퍼스 공학원 447
Lab: LangAGI Lab
Email: jinyeo@yonsei.ac.kr
Research
AI Agents
Long-horizon agents, world models, tool use, and interactive learning.
Human–AI Collaboration
Collaborative decision making, behavior alignment, and clinical agents.
Medical AI
Multimodal clinical reasoning, ophthalmic imaging, and decision support.
News
- Research Collaboration with Samsung Electronics DX
- Lecture at SK Group ML Engineer Program
- Track Chair at Global AI Frontiers Symposium
- Talk at LG Electronics CTO
- Talk at Samsung Global Research
- Talk at Samsung Display Research
Publications
2026
-
COLM 2026
Scalable workflow-grounded environments and reward designs for training procedurally compliant language agents.
-
On Training Large Language Models for Long-Horizon Tasks: An Empirical Study of Horizon Length
ICML 2026 Featured in DAIR.AI (Top AI Paper of the Week)
Studies how task horizon length affects language-model training and performance on long-horizon tasks.
-
EMBGUARD: Constructing Hazard-Aware Guardrails for Safe Planning in Embodied Agents
ICML 2026
Introduces hazard-aware guardrails for safer planning and action selection in embodied agents.
-
PAC-BENCH: Evaluating Multi-Agent Collaboration under Privacy Constraints
ACL 2026 Findings
Benchmarks how multiple agents collaborate while respecting privacy constraints.
-
Why These Documents? Explainable Generative Retrieval with Hierarchical Category Paths
ACL 2026 Findings
Makes generative retrieval more interpretable through hierarchical category-path explanations.
-
MVIGER: Multi-View Variational Integration of Complementary Knowledge for Generative Recommender
SIGIR 2026
Combines complementary knowledge views through variational integration for generative recommendation.
-
International Journal of Medical Informatics, 2026 (and Asia Retina Congress 2025) Best Poster Award
Analyzes when human–AI collaboration succeeds or fails on difficult real-world clinical reasoning cases.
-
ICLR 2026
Examines personalization in embodied agents through the challenges and opportunities of memory utilization.
-
Prior-based Noisy Text Data Filtering: Fast and Strong Alternative For Perplexity
ICLR 2026
Proposes a fast prior-based method for filtering noisy text as an alternative to perplexity-based selection.
-
Quantifying Genuine Awareness in Hallucination Prediction: Disentangling Question-Side Shortcuts
Agentic AI in the Wild Workshop, ICLR 2026
Separates genuine hallucination awareness from shortcuts that rely only on properties of the question.
-
AgenticShop: Benchmarking Agentic Product Curation for Personalized Web Shopping
WWW 2026
Introduces a benchmark for personalized product curation by web-shopping agents.
2025
-
Web-Shepherd: Advancing PRMs for Reinforcing Web Agents
NeurIPS 2025 Spotlight
Develops process reward models that better reinforce web agents during multistep interaction.
-
Fast and Fluent Diffusion Language Models via Convolutional Decoding and Rejective Fine-tuning
NeurIPS 2025 Spotlight
Improves the speed and fluency of diffusion language models with convolutional decoding and rejective fine-tuning.
-
ToolHaystack: Stress-Testing Tool-Augmented Language Models in Realistic Long-Term Interactions
EMNLP 2025 Findings
Stress-tests tool-augmented language models in realistic, long-term tool-use interactions.
-
PRINCIPLES: Synthetic Strategy Memory for Proactive Dialogue Agents
EMNLP 2025 Findings
Uses synthetic strategy memories to make dialogue agents more proactive and consistent.
-
EMNLP 2025 Findings
Evaluates conversational recommenders with target-free user simulation rather than hidden target guessing.
-
EMNLP 2025 Findings
Investigates whether English–Korean code-switching activates distinct knowledge behavior in language models.
-
Designing Memory-Augmented AR Agents for Spatiotemporal Reasoning in Personalized Task Assistance
XRMemory Workshop, IEEE ISMAR 2025
Designs memory-augmented augmented-reality agents for personalized spatiotemporal task assistance.
-
Rethinking Reward Model Evaluation Through the Lens of Reward Overoptimization
ACL 2025
Reframes reward-model evaluation around robustness to reward overoptimization.
-
ACL 2025
Benchmarks whether large language models can understand and generate structured scene graphs.
-
Can You Share Your Story? Modeling Clients' Metacognition and Openness for LLM Therapist Evaluation
ACL 2025 Findings
Models client metacognition and openness to improve evaluation of LLM-based therapeutic conversations.
-
ACL 2025 Industry
Provides data for mitigating the cold-start problem when training short chain-of-thought reasoning models with reinforcement learning.
-
Ophthalmology Science, 2025
Studies age-related hypofluorescent spots as a prognostic factor in polypoidal choroidal vasculopathy.
-
Review-driven Personalized Preference Reasoning with Large Language Models for Recommendation
SIGIR 2025
Uses review evidence to improve personalized preference reasoning for recommendation.
-
Web Agents with World Models: Learning and Leveraging Environment Dynamics in Web Navigation
ICLR 2025
Learns and exploits environment dynamics to improve long-horizon web navigation.
-
Towards Lifelong Dialogue Agents via Timeline-based Memory Management
NAACL 2025
Manages conversational memories along a timeline to support lifelong dialogue agents.
-
NAACL 2025 Findings
Introduces a psychometric test set for measuring personality consistency in large language models.
-
Nature Scientific Reports, 2025
Predicts future branch retinal vein occlusion from multimodal pre-onset fundus images.
-
WSDM 2025
Aligns cross-lingual entities without supervision by matching neighboring triples and textual information.
2024
-
Train-Attention: Meta-Learning Where to Focus in Continual Knowledge Learning
NeurIPS 2024
Meta-learns where a model should focus when continually acquiring new knowledge.
-
EMNLP 2024 Featured in Hugging Face (Daily Papers)
Uses pseudocode execution as an intermediate reasoning process for algorithmic problem solving.
-
COFEE-GYM: An Environment for Evaluating and Improving Natural Language Feedback on Erroneous Code
EMNLP 2024
Provides an environment for evaluating and improving natural-language feedback on erroneous code.
-
Evidence-Focused Fact Summarization for Knowledge-Augmented Zero-shot Question Answering
EMNLP 2024
Summarizes evidence relevant to knowledge-augmented zero-shot question answering.
-
CACTUS: Towards Psychological Counseling Conversations using Cognitive Behavioral Theory
EMNLP 2024 Findings
Builds psychologically grounded counseling conversations using principles from cognitive behavioral theory.
-
EMNLP 2024 Findings
Uses question-driven reasoning to identify implicit knowledge for more insightful table summaries.
-
ACL 2024 Outstanding Paper Award
Reduces preference bias to improve emotional-support conversations with large language models.
-
VERIFINER: Verification-augmented NER via Knowledge-grounded Reasoning with Large Language Models
ACL 2024
Augments named-entity recognition with knowledge-grounded reasoning and verification.
-
PEARL: A Review-driven Persona-Knowledge grounded Conversational Recommendation Dataset
ACL 2024 Findings
Introduces a review-driven persona- and knowledge-grounded dataset for conversational recommendation.
-
Self-Consistent Reasoning-based Aspect-Sentiment Quad Prediction with Extract-then-Assign Strategy
ACL 2024 Findings
Applies self-consistent extract-then-assign reasoning to aspect–sentiment quad prediction.
-
RTSUM: Relation Triple-based Interpretable Summarization with Multi-level Salience Visualization
NAACL 2024 Demonstration
Uses question-driven reasoning to identify implicit knowledge for more insightful table summaries.
-
EACL 2024 Best Paper Award in KCC 2023 (Preliminary study)
Refines persona information with commonsense to improve long-term conversational memory.
-
Evidentiality-Aware Retrieval for Overcoming Abstractiveness in Open-Domain Question Answering
EACL 2024 Findings
Uses evidentiality-aware retrieval to address abstractiveness in open-domain question answering.
-
Investigative Ophthalmology & Visual Science (IOVS), 2024
Uses multitask learning to distinguish necrotizing viral from non-infectious retinitis using blood and serology data.
-
AAAI 2024
Introduces reasoning-aware diagnostic prompting with model-generated clinical rationales.
2023
-
Dialogue Chain-of-Thought Distillation for Commonsense-aware Conversational Agents
EMNLP 2023
Distills dialogue chain-of-thought reasoning into commonsense-aware conversational agents.
-
European Radiology, 2023
Predicts glioma IDH genotype from free-text MR radiology reports using natural language processing.
-
CoTEVer: Chain of Thought Prompting Annotation Toolkit for Explanation Verification
EACL 2023 Demonstration
Provides an annotation toolkit for verifying chain-of-thought explanations.
-
Evidence-Empowered Transfer Learning for Alzheimer's Disease
IEEE International Symposium on Biomedical Imaging (ISBI), 2023
Transfers evidence across tasks to improve learning for Alzheimer’s disease prediction.
-
TUTORING: Instruction-grounded Conversational Agent for Language Learners
AAAI 2023 Demonstration
Builds an instruction-grounded conversational tutor for language learners.
2022
-
EMNLP 2022
Automatically curates large-scale multi-skill dialogue datasets from machine-generated conversations.
-
Mind the Gap! Injecting Commonsense Knowledge for Abstractive Dialogue Summarization
COLING 2022
Injects commonsense knowledge to improve abstractive dialogue summarization.
-
Modularized Transfer Learning with Multiple Knowledge Graphs for Zero-shot Commonsense Reasoning
NAACL 2022
Transfers knowledge from multiple graphs for zero-shot commonsense reasoning.
-
Dual Task Framework for Improving Persona-grounded Dialogue Dataset
AAAI 2022
Improves persona-grounded dialogue data with a dual-task learning framework.
-
TrustAL: Trustworthy Active Learning using Knowledge Distillation
AAAI 2022
Combines active learning and knowledge distillation for more trustworthy sample selection.
-
Nature Scientific Reports, 2022
Predicts photodynamic-therapy outcomes for chronic central serous chorioretinopathy using multimodal transfer learning.
~2021
-
Label and Context Augmentation for Response Selection at DSTC8
IEEE/ACM TASLP, 2021
Augments labels and conversational context to improve response selection.
-
Less is More: Attention Supervision with Counterfactuals for Text Classification
EMNLP 2020
Uses counterfactual attention supervision to improve text classification with limited annotation.
-
Conversion Prediction from Clickstream: Modeling Market Prediction and Customer Predictability
IEEE TKDE, 2020
Models market-level and customer-level predictability for online purchase conversion.
-
XINA: Explainable Instance Alignment using Dominance Relationship
IEEE TKDE, 2020
Explains instance alignment through dominance relationships between candidate matches.
-
Learning with Limited Data for Multilingual Reading Comprehension
EMNLP 2019
Transfers limited supervision across languages for multilingual reading comprehension.
-
Soft Representation Learning for Sparse Transfer
ACL 2019
Learns soft representations that improve transfer under sparse supervision.
-
Translations as Additional Contexts for Sentence Classification
IJCAI 2018
Uses machine translations as auxiliary contexts for sentence classification.
-
Visual Choice of Plausible Alternatives: An Evaluation of Image-based Commonsense Causal Reasoning
LREC 2018
Evaluates image-based commonsense causal reasoning through plausible alternative selection.
-
Machine-translated Knowledge Transfer for Commonsense Causal Reasoning
AAAI 2018
Transfers machine-translated knowledge for commonsense causal reasoning.
-
Efficient Keyword-aware Representative Travel Route Recommendation
IEEE TKDE, 2017
Recommends representative travel routes while accounting for user-specified keywords.
-
Multimodal KB Harvesting for Emerging Spatial Entities
IEEE TKDE, 2017
Harvests knowledge about emerging spatial entities from multimodal sources.
-
Predicting Online Purchase Conversion for Retargeting
WSDM 2017
Predicts purchase conversion to support personalized retargeting.
-
Event Grounding from Multimodal Social Network Fusion
ICDM 2016
Grounds events by fusing multimodal evidence from social networks.
-
Browsing2purchase: Online Customer Model for Sales Forecasting in an E-Commerce Site
WWW 2016
Models browsing behavior to forecast purchases in e-commerce.
-
Purchase Influence Mining: Identifying Top-k Items Attracting Purchase of Target Item
WWW 2016
Identifies products that most strongly influence purchases of a target item.
-
Understanding Emerging Spatial Entities
AAAI 2016
Models and understands newly emerging spatial entities from web data.
-
KSTR: Keyword-aware Skyline Travel Route Recommendation
ICDM 2015
Recommends keyword-aware skyline travel routes under multiple user preferences.
-
Finding Influential Products on Social Domination Game
CIKM 2012
Identifies influential products using a social-domination formulation.
Teaching
- [AAI5003] Deep Learning for Natural Language Processing
- [CAS3301] Natural Language Processing
- [CAS2250] Introduction to Artificial Intelligence
Students
I lead the LangAGI Lab at Yonsei University. We study language models, agents, interaction, and medical AI. Prospective students may contact me with a CV and a brief description of their research interests.
Short Bio
I am an Associate Professor in the Department of Artificial Intelligence at Yonsei University. Before joining Yonsei University, I worked at SKT Brain. I received my Ph.D. from POSTECH and completed research internships at Adobe Research, San Jose.