Publications

To improve the performance of a database, database parameter tuning is required. Existing machine learning-based tuning models fail to sufficiently capture the complex dependency relationships among parameters. This paper proposes PDTune, a model that leverages a Large Language Model (LLM) with rich domain knowledge to identify interdependencies among database parameters during the optimization process and incorporates them into the search direction to enhance exploration efficiency. Invited paper (KCC 2025 Outstanding Presentation).

Jeju experiences over 10 million annual tourist visits, generating 3,000-4,000 pothole complaints yearly. This study proposes an interpretable ML framework combining machine learning with SHAP (SHapley Additive exPlanations) to predict road maintenance intervention without requiring direct pavement condition data. Using only annual traffic, weather, and road characteristics, the model identifies high-likelihood maintenance links and quantifies nonlinear variable interactions (notably summer heat × heavy vehicle traffic), providing a scientific basis for proactive municipal budget allocation.

Potholes accelerate pavement deterioration and threaten traffic safety, while traditional manual maintenance suffers from reporting-and-dispatch delays. This paper proposes a fully automated framework integrating crowdsourced real-time detection with robotic repair. Applying a one-dimensional corridor model to 15 major Jeju Island routes quantifies total delay reductions: detection-to-repair time is cut by more than 90% and administrative wait time is effectively eliminated, demonstrating the technical and economic feasibility of data-driven pavement management.

Vulnerable pedestrians experience reduced walking speed and elevated accident risk at intersections due to various health conditions. This paper proposes a real-time crosswalk safety monitoring system that estimates pedestrian movement speed and predicts estimated time of arrival (ETA) from CCTV footage. A YOLOv8 model trained on AI-Hub data is used to detect and track pedestrians in complex intersection environments, enabling proactive signal control and risk alerting.

The purpose of this study was to examine the impact of ESG rating changes of companies listed in the Korean Stock Exchange on stock returns. Based on yearly changes of ESG ratings we grouped companies into two portfolios (upgrade and downgrade) and calculated portfolios' returns. In the COVID-19 period (2020–2021), the upgraded portfolio outperforms the downgraded portfolio by 0.7 percentage points per month, statistically significant after controlling for Carhart four factors. This study is the first to examine the impact of ESG rating changes on stock returns in Korea.

Text mining or natural language processing is a multi-discipline technique that can distill and obtain useful information from text. With the development of AI and increasing abundance of big data, text mining is becoming one of the essential technologies in the fields of business, government, and academia. This paper provides a detailed description of how text data analysis can be used for economic analysis by reviewing key methodologies and related literature. We expect that text mining will complement the limitations of current data and methodologies by serving as a new source of information.

This study investigates the existence of additional information in the monetary policy board (MPB) minutes of the Bank of Korea (BOK) through topic modeling and sentiment analysis. We extract five topics: the economy, monetary policy, financial market, prices, and debt. An ordered probit model reveals that tones of monetary policy, prices, and debt have additional information on the current interest rate decision, and the tone of the economy has predictive power for the future interest rate decision.

We propose a novel approach to measure monetary policy shocks using sentiment analysis, which is relatively free from specification errors compared to VAR-identified shocks. We quantify the tones of news articles around 152 dates of Monetary Policy Board (MPB) meetings of the Bank of Korea (BOK) from 2005 to 2017 and measure monetary policy surprises using changes in those tones following monetary policy announcements. Our text-based measure better explains changes in long-term rates, while changes in the base rate and VAR-identified shocks are more closely associated with changes in short-term rates.