CV
Curriculum Vitae
Contact Information
| Name | Seungyeon Back |
| Professional Title | M.S. Student in Artificial Intelligence, Hanyang University | Visiting Student, Carnegie Mellon University |
| seungyeo@andrew.cmu.edu | |
| Location | 5000 Forbes Avenue, Pittsburgh, PA 15213 |
Professional Summary
M.S. student in Artificial Intelligence at Hanyang University, advised by Prof. Dong-Kyu Chae, with a B.S. in Mathematics, currently a visiting student at Carnegie Mellon University. Research interests include LLM Reasoning, LLM Agent, and RAG. Co-first author of AdapDict (CSCW 2026) and currently working on graph-based retrieval for real-time recommendation in collaboration with LG Electronics.
Experience
-
2026 - Present Seoul, South Korea
Graduate Researcher, GraphRAG-based Recommendation System
Hanyang University (in collaboration with LG Electronics)
Building a graph-based retrieval and reasoning pipeline for real-time content recommendation.
- Built a real-time data pipeline for news and OTT content ingestion and processing.
- Designed a dual-graph framework aligning news–content graphs via shared entities.
- Developed a graph-based retrieval pipeline using multi-hop traversal to generate candidate content.
- Integrated LLMs for path-based reasoning over graph structures, enabling explainable recommendations.
-
2023 - 2024 Seoul, South Korea
Undergraduate Researcher
Applied Statistics Lab, Hanyang University
- Implemented LSTM, GRU, and BERT-based models for stock prediction using PyTorch, incorporating technical indicators and news data.
- Participated in collaborative study sessions on time-series modeling and machine learning, including peer-led presentations and discussions.
Education
-
2025 - Present Seoul, South Korea
M.S.
Hanyang University
Artificial Intelligence
- Advised by Prof. Dong-Kyu Chae
- Research interests: LLM Reasoning, LLM Agent, RAG
-
2026 - 2027 Pittsburgh, PA, USA
-
2021 - 2025 Seoul, South Korea
Publications
-
2026 AdapDict: A Level- and Domain-Adaptive Educational Dictionary and Encyclopedia System
CSCW 2026 (Accepted)
Proposed a level- and domain-adaptive explanation framework for controllable explanation generation in LLMs. Designed a generator-verifier loop with structured outputs, reducing hallucination by 50%. Developed a user-adaptive prompting mechanism for multi-dimensional control of explanation difficulty. Conducted a user study (n=41) demonstrating improved clarity, usability, and domain-specific understanding over baseline tools. (Co-first author)
-
2026 An Empirical Study of the Clustering of Mountain Hiking GPS Trajectory Data
Under Revision, IEEE Access
Proposed a novel trajectory similarity measure integrating spatial, temporal, and velocity components for irregular spatio-temporal data. Designed an epsilon-ball-based spatial similarity metric and a data-adaptive weighting framework (alpha, beta, gamma) for flexible similarity modeling. Applied hierarchical clustering to large-scale GPS trajectory data to uncover latent behavioral patterns. (Co-first author)