SIAI Yearbook – 2023
For the first time, SIAI students had the opportunity to present their research papers. The 2023 papers can be summarized under the theme “The Use of AI Algorithms in Corporate Management”. With the rise of ChatGPT, interest in how companies can leverage AI algorithms has surged, and the students have addressed this through their individual research topics. The Yearbook will provide commentary on the papers and share the students’ experiences during the process of writing their paper.
Researcher Jeonghoon Song discussed the topic of “Modeling Joint Distribution Of Monthly Energy Uses In Individual Urban Buildings For A Year”. Monthly energy usage data from individual buildings is crucial for estimating energy consumption in urban areas. Previous studies based on regression analysis introduced models that estimate the mean and variance of monthly energy usage. However, these models did not account for the correlation between energy usage across different months, leading to difficulties in predicting the dependent variable. Therefore, this study aimed to provide a more realistic prediction of monthly energy usage, by building and energy type (electricity/gas), by reflecting the correlation between months.
Researcher Hyeyoung Park discussed the topic, “Is the bubble in the housing auction market really a bubble?” She investigated the real estate bubble through data from the Gangnam apartment auction market, analyzing the price difference between the winning bidder and the runner-up to identify and verify elements contributing to the “housing bubble”. The research aimed to determine whether the “winner’s curse” truly corresponds to a bubble.
Researcher Bohyun Yoo analyzed the characteristics of the real estate auction market with the topic “Price Premium Discovery In Real Estate Auction Market”. He addressed existing limitations caused by differences between the court appraisals and auction closing prices, as well as the difficulty in incorporating external data for court appraisals and discount/premium factors. Fourier transform was used to inversely extract the impact of court appraisals and discount/premium factors.
Researcher Mincheol Kim presented on the topic “Interpretable Topic Analysis,” building upon Andrew Ng’s LDA model by utilizing a logit-normal distribution. The logit-normal distribution maintains multi-modality while being differentiable, allowing for faster computation. This method was used for big data analysis of keywords from newspaper articles.
Researcher Jeongwoo Park focused on “Modeling Digital Advertising Data with Measurement Error”. Measuring the effect of digital advertising is challenging due to the presence of endogenous factors such as measurement errors. This analysis used a Poisson distribution-based time series analysis, confirming that in the short term, the Kalman filter model is more suitable, while in the long term, a data-dependent Poisson time series model is more appropriate. Ultimately, an ensemble model combining the two approaches was proposed.