The Computational Linguistics and Knowledge (CLinK) Lab at Pusan National University (PNU) studies how to turn language into knowledge: we extract facts and relations from text, organize them into knowledge bases and graphs, and build and evaluate language models that use them. We work in domains where expert knowledge matters, including biomedicine and nutrition, food and cooking, finance, public safety, and industry.
Figure: RECoDe, Bioinformatics 2026 (CC BY 4.0).
We extract entities and relations from biomedical literature and organize them into knowledge graphs, with a focus on diet, non-communicable diseases, and biomarkers. Much of this work was done in the EU CoDiet consortium, and we now use it to check whether advice from large language models (LLMs) on medical diets agrees with the published evidence.
Key work: RECoDe (Bioinformatics, ECCB 2026), Evidence-based knowledge graph (LMRL Workshop at ICLR 2025), CoDiet corpus (Scientific Data, in press; preprint).
Projects: Trustworthy medical-dietary LLMs (MSIT / NRF, 2026-2029); CoDiet (EU Horizon Europe and UKRI, 2023-2026).
Figure: CookingSense, LREC-COLING 2024 (CC BY-NC 4.0).
We study recipes and culinary knowledge with machine learning and language models. Topics include food pairing and recipe completion, ingredient quantity prediction, a culinary knowledge base and benchmark for food decision support, recipe idea generation, and the evaluation of how LLMs adapt dishes to other cuisines.
Key work: RecipeMind (CIKM 2022), KitchenScale (Expert Systems with Applications, 2023), CookingSense (LREC-COLING 2024), ASH (IEEE Access, 2026).
Projects: Personalized nutrition AI for clinical nutrition education (PNU, 2026); Collaboration with Sony AI's Gastronomy Flagship Project (Sony AI, 2021-2023).
Figure: SciNumQA, IEEE Access 2026 (CC BY-NC-ND 4.0).
We build benchmarks and datasets for language models and test whether evaluation results can be trusted, especially for Korean. Topics include LLM-as-a-judge reliability, real Korean user questions (NativeKo-Arena), mixed sentiment in Korean sentences, reasoning inconsistency on medical questions, cross-lingual consistency of medical vision-language models, and numerical reasoning over scientific tables.
Key work: Korean MT-Bench, Beyond Accuracy, and Korean mixed sentiment dataset (all Database Research, 2026, in Korean); SciNumQA (IEEE Access, 2026).
Projects: Trustworthy medical-dietary LLMs (MSIT / NRF, 2026-2029); PNU 2-Year Research Grant (2025-2027).
Figure: Counter-Scam, IJCAI 2026 (c) IJCAI.
We build language model systems that support professionals by combining retrieval, multi-agent workflows, and small fine-tuned models, with applications in police investigation, scam response, and medicine. Current projects extend this to Korean small language models for regional industries, document-grounded question answering, and multi-agent support for shipbuilding production.
Key work (co-authored): LAPIS (CIKM 2024), Counter-Scam (IJCAI 2026), Meerkat (npj Digital Medicine, 2025).
Projects: Shipbuilding production knowledge (MOTIR / KEIT, 2026-2028); Document-based retrieval-augmented generation (RAG) for domain knowledge search (MSIT / Busan Metropolitan City, 2026); Korean small language models for regional industries (PNU, 2026-2027).
Figure adapted from DeepAries, CIKM 2025 (CC BY-NC-SA 4.0).
We apply deep reinforcement learning to portfolio selection. Topics include hierarchical allocation across asset classes, portfolio decisions that use market forecasts, and learning when to rebalance as well as how to allocate.
Key work: HADAPS (IEEE Access, 2023), DeepClair (CIKM 2024), DeepAries (CIKM 2025).
Trustworthiness Verification of Medical-Dietary Decision-Making LLMs through Disease-Diet Knowledge Graph and Automated Systematic Review with Application to Nutrition and Cooking Domains, MSIT / NRF (Core Research Grants, General Research A), 2026-2029. PI. Project No. RS-2026-25604981.
We build a disease-diet knowledge graph that records the amount and strength of evidence behind each relation; design constrained-output reasoning that grounds LLM answers in that evidence, the user's needs, and the current context; and compare the automated results with expert systematic reviews.
Development of Knowledge-Datafied Production Support Technology for Shipbuilding, MOTIR / KEIT (Machinery and Equipment Industrial Technology Development Program), 2026-2028. PI (PNU). Project No. RS-2026-25608503.
We build the orchestration layer of the multi-agent system: a supervisor agent interprets complex questions from shipyard workers and managers, routes them to specialized agents that handle production knowledge and data, and returns answers with their supporting evidence. Field feedback is used to improve the system.
Development of Korean Small Language Models (1B Parameters or Fewer) Specialized for Regional Industries and a Synthetic-Corpus-Based Training and Education Pipeline, Pusan National University (AI Teaching Innovation Model Research Program), 2026-2027. PI.
Development of Core Technologies for a Document-Based RAG Pipeline and Conversational AI Platform for Domain-Specific Knowledge Search, MSIT / Busan Metropolitan City (Busan Research Industry Promotion Complex Development Program), 2026. PI.
With an industry partner, we build the retrieval core of a conversational AI platform for Korean domain documents such as regulations and manuals: hybrid keyword and vector search over text, images, and tables, and answers that cite their sources or abstain when the evidence is insufficient.
Private Project (Food and Cooking AI), industry-sponsored, 2026. PI.
Development of a Continuous Glucose Monitoring (CGM)-Based Personalized Nutrition AI Learning Support System and an Empirical Study of an Innovative Clinical Nutrition Education Model, Pusan National University (AI Teaching Innovation Model Research Program), 2026. Co-investigator.
AI Star Fellowship Support Program (Pusan National Univ., Regional University-led), MSIT / IITP, 2026-2031. Participating researcher. Project No. RS-2026-25615271.
Develops next-generation multimodal medical foundation models and a precision-medicine platform with clinical and industry partners, and trains top-tier early-career AI researchers in the region.
PNU AI Convergence Graduate School: Artificial Intelligence Convergence Innovation Human Resources Development (Pusan National University), MSIT / IITP, since 2025. Participating faculty. Project No. RS-2023-00254177.
Trains AI talent for the digital transformation of the southeast region through industry-academia research projects in manufacturing, logistics, and healthcare.
Innovative Human Resource Development for Local Intellectualization (Pusan National University), MSIT / IITP, since 2026. Participating faculty. Project No. RS-2026-25590495.
Trains AI talent for the AI transformation of manufacturing and port industries in the southeast region through joint research with partner companies, a degree program for working professionals, and regional cooperation.
PNU AI Bootcamp, MOE / KIAT (Advanced Industry Talent Development Bootcamp, AI field), since 2026. Co-lead of the Convergence AI Team and lead of the Generative AI track.
PNU Software-Centered University (Phase 2), MSIT / IITP (National Program of Excellence in Software), since 2026. Co-lead of the Software Convergence Education Division.
Korean Southeast Center for the 4th Industrial Revolution Leader Education, MOE / NRF (BK21 FOUR), since 2026. Participating faculty.
Private Project (Manufacturing Knowledge Search), industry collaboration, 2026. PI.
Data-driven Medical Technology Development and Advancement based on Artificial Intelligence and Big Data, MOHW / KHIDI (Research-Driven Hospital Program), 2020-2028. Researcher at Korea University, 2025. Project No. RS-2020-KH088565.
Develops AI and big-data methods for drug discovery, molecular diagnostics, and medical question answering; the work includes a biomedical knowledge graph built from the literature and a question-answering system for medical professionals.
Developing Human-in-the-loop Large Language Model for Biomedical Hypothesis Co-Design, MSIT / NRF (Mid-Career Researcher Program), 2023-2028. Researcher at Korea University, 2025. Project No. RS-2023-NR077250.
Develops large language models that generate biomedical hypotheses from the literature together with human experts.
CoDiet: Combatting Diet Related Non-Communicable Disease through Enhanced Surveillance, European Union and UKRI (Horizon Europe), 2023-2026. Research Associate at Imperial College London, 2023-2025 (Work Package 1: text corpus, relation extraction, and knowledge graph).
Collaboration with Sony AI's Gastronomy Flagship Project, Sony AI (research contracts with Korea University), 2021-2023. Researcher at Korea University and Research Intern at Sony AI.
Abbreviations: IITP, Institute of Information & Communications Technology Planning & Evaluation; KEIT, Korea Planning & Evaluation Institute of Industrial Technology; KHIDI, Korea Health Industry Development Institute; KIAT, Korea Institute for Advancement of Technology; MOE, Ministry of Education; MOHW, Ministry of Health and Welfare; MOTIR, Ministry of Trade, Industry and Resources; MSIT, Ministry of Science and ICT; NRF, National Research Foundation of Korea; PI, principal investigator; UKRI, UK Research and Innovation. Project numbers are Korean national R&D project numbers registered in IRIS (Integrated R&D Information System).