Seoul, South Korea
Beyond Ontology in Dialogue State Tracking (Mar 2024 – Aug 2024): Designed a framework combining LLM and VGAE to overcome limitations of ontology-based dialogue state tracking. Achieved SOTA accuracy without relying on predefined ontologies. Accepted at ICKG 2024. Led the entire project as the team leader. Text Mining-based Microbiology Paper Exploration System (May 2023 – Mar 2024): Developed a microbiology-specialized language model with NER, RE, and keyword extraction capabilities. Enhanced literature search efficiency through dual semantic-symbolic retrieval and knowledge graph visualization. Applied in the research environment of KRIBB (Korea Research Institute of Bioscience and Biotechnology); registered for copyright. AI Strategy for Intelligent Digital Library (Apr 2023 – Jul 2023): Before the popularization of LLMs, conducted user research at the National Assembly Library and proposed an AI adoption strategy. Designed and recommended the integration of a RAG (Retrieval-Augmented Generation) architecture. Text Mining-based Cell Line Paper Exploration System (Mar 2023 – Jun 2023): Built a cell line-specific language model with NER, RE, and keyword extraction. Developed a semantic search system by paper section and visualized knowledge graphs. Presented at DTMBIO 2023; registered for copyright.
Korean Dialogue Context Inference Challenge by the National Institute of the Korean Language (Jul 2024 – Aug 2024): Applied strategies such as AI-based automatic prompt generation, quantization, and ORPO ensemble to improve efficiency and accuracy in Korean utterance type classification. Achieved 2nd place on the official leaderboard despite limited training resources. Led the entire project as the team leader. Prompt-Based Chatbot for Essay and Interview Preparation (Apr 2024 – Jun 2024): Ensured diversity and logical consistency of generated questions using LLM-based prompt strategies and data augmentation. Took charge of modeling and project management. Multilingual Hierarchical E-commerce Product Classification (Oct 2023 – Dec 2023) (in collaboration with Lether AI): Implemented a multilingual hierarchical classification model by applying hyperbolic embeddings and hierarchy-preserving loss to XLM-RoBERTa, based on the HiTIN paper. Improved both domain generalization and per-level accuracy. Contributed as a team member to paper implementation and task conversion. Korean Contextual Profanity Detection Model (Aug 2023): Collected and refined large-scale user comment data to detect context-dependent Korean profanity. Developed a context-aware classifier using KoBERT. Led the entire process from ideation to implementation.
- Reference Service - Faculty support - Cataloguing, Classification