http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
지방자치단체 주요재정사업 성과와 예산집행 변수 간 관계성 연구: 서울시 주요재정사업평가 결과를 바탕으로
공준석(Kong, Joon Suk) 서울행정학회 2024 한국사회와 행정연구 Vol.34 No.4
본 연구는 예산집행 관련 변수가 사업성과에 미치는 영향을 파악하기 위해, 서울시의 일별 지출 자료 약 38만 건과 주요재정사업평가 결과 약 600건을 연결하여, 사업별 일별 지출과 주요재정사업평가 결과의 관계를 분석하였다. 이를 위해 일별 지출 자료에서 총집행액, 집행횟수, 집행기간, 상반기 집행률 및 집행속도지수 변수를 생성하여, 주요재정상업평가 결과(등급)에 대한 ANOVA 분석과 평가 결과에 대한 다중회귀분석을 실시하였다. ANOVA 분석결과 집행횟수가 5회 이상인 경우 선행연구와 같이 주요재정사업평가 등급에 따라 총집행액의 평균차이가 존재하며, 다중회귀분석 결과 집행횟수, 상반기 집행률 및 집행속도는 평가등급에 다소간의 영향력이 있는 것으로 나타났다. 본 연구를 통해 예산집행에 대한 정책적 방향성을 제시하고, 집행에 관련된 변수를 확장하는데 기여하기를 기대한다. This study examines the influence of budget execution factors on project performance in Seoul. It links around 380,000 daily expenditure records with about 600 financial project evaluations. Key variables include total execution amount, execution count, duration, first-half execution rate, and speed index. Using ANOVA and multiple regression analyses, it finds that while total execution amount varies by evaluation grade, first-half execution rate shows no significant difference. However, the regression analysis suggests that modifying the number of executions, first-half rate, and speed can improve performance. This research offers policy guidance on budget execution and broadens the scope of related variables.
딥러닝 기법을 이용한 공공기관 설문조사 주관식 문항 활용방안에 관한 연구
공준석(Joon Suk Kong),김유영(Yu Yeong Kim) 한국자료분석학회 2023 Journal of the Korean Data Analysis Society Vol.25 No.6
본 연구의 목적은 공공기관 설문조사에서 주관식 문항의 포함 여부가 만족도 조사 결과에 미치는 영향을 딥러닝 기법을 통해 분석하는 데 있다. 매년 약 1,000여 개의 공공기관이 만족도 조사를 시행하고 있으나, 비교가능성을 위해 보편화된 설문을 사용함에 따라 구체적 정보의 확인에 한계를 보이며, 이에 대한 대안으로 주관식 문항이 활용되고 있다. 그러나 주관식 문항에 대한 분석은 분절적으로 이루어져, 실질적으로 만족도 조사에 활용되지 못하고 있다. 이와 같은 문제를 해결하기 위해 K-공공기관 만족도 조사 결과 약 1,500명의 데이터를 통해 주관식 문항의 활용방안을 탐색했다. RoBERTa 알고리즘을 통해 주관식 문항을 Embedding하고, XGBoost, 인공신경망, 로지스틱 회귀분석 등의 알고리즘을 활용하여 만족도 예측 결과를 비교했다. 실험 결과, 알고리즘에 따른 성능 차이는 일반적으로 알려진 것과 같이 Table data에서는 XGBoost의 성능이 우수한 것으로 나타났다. 하지만 주관식 문항을 포함할 때는 결과가 상이할 수 있음을 확인하였다. 나아가 주관식 문항을 포함하여 만족도 점수를 예측할 때 정확도(accuracy)와 재현률(recall)에 긍정적인 영향을 미칠 수 있음을 확인하였다. 따라서 딥러닝 기반의 분석 방법을 통해 주관식 문항의 복잡성과 다양성을 더욱 정확하게 반영할 수 있으며, 이를 통해 만족도 조사의 효과성과 투명성을 높일 수 있다고 기대된다. The objective of this study is to assess the impact of incorporating subjective questions in public agency satisfaction surveys using deep learning techniques. Each year, around 1,000 public agencies carry out satisfaction surveys. Nevertheless, the reliance on standardized surveys for comparability restricts the gathering of nuanced information. In response, subjective questions have been introduced. Yet, the examination of these questions remains fragmented and hasn't been integrated into mainline satisfaction surveys. To address this gap, we evaluated feedback from roughly 1,500 participants associated with K-Public Agency. For handling these subjective questions, we leveraged the RoBERTa algorithm and employed a range of models, such as XGBoost, neural networks, and logistic regression. Preliminary findings underscore the effectiveness of XGBoost when dealing with structured data. Notably, its performance showed variations when considering subjective inputs. A significant finding was that integrating subjective questions enhanced both the accuracy and recall in predicting satisfaction levels. This indicates that deep learning offers a detailed interpretation of subjective inquiries, which elevates the overall quality of public satisfaction surveys.
탑승자 교통사고에서 경추손상 판단을 위한 중증도 요인 분석
이희영,육현,공준석,강찬영,성실,이정훈,김호중,김상철,추연일,전혁진,박종찬,최지훈,이강현,Lee, Hee Young,Youk, Hyun,Kong, Joon Seok,Kang, Chan Young,Sung, Sil,Lee, Jung Hun,Kim, Ho Jung,Kim, Sang Chul,Choo, Yeon Il,Jeon, Hyeok Jin,Park, Jon 한국자동차안전학회 2018 자동차안전학회지 Vol.10 No.3
It was a pilot study for developing an algorithm to determine the presence or absence of cervical spine injury by analyzing the severity factor of the patients in motor vehicle occupant accidents. From August 2012 to October 2016, we used the KIDAS database, called as Korean In-Depth Accident Study database, collected from three regional emergency centers. We analyzed the general characteristics with several factors. Moreover, cervical spine injury patients were divided into two groups: Group 1 for from Quebec Task Force (hereinafter 'QTF') grade 0 to 1, and group 2 for from QTF grade 2 to 4. The score was assigned according to the distribution ratio of cervical spine injured patients compared to the total injured patients, and the cut-off value was derived from the total score by summation of the assigned score of each factors. 987 patients (53.0%) had no cervical spine injuries and 874 patients (47.0%) had cervical spine injuries. QTF grade 2 was found in 171 patients (9.2%) with musculoskeletal pain, QTF grade 3 was found in 38 patients (2.0%) with spinal cord injuries, and QTF grade 4 was found in 119 patients (6.4%) with dislocation or fracture, respectively. We selected the statistically significant factors, which could be affected the cervical spine injury, like the collision direction, the seating position, the deformation extent, the vehicle type and the frontal airbag deployment. Total score, summation of the assigned each factors, 10 was presented as a cut-off value to determine the cervical spine injury. In this study, it was meaningful as a pilot study to develop algorithms by selecting limited influence factors and proposing cut-off value to determine cervical spine injury. However, since the number of data samples was too small, additional data collection and influencing factor analysis should be performed to develop a more delicate algorithm.
한국형 실사고 심층조사 데이터베이스 질향상을 위한 차량속도(ΔV) 측정방법에 관한 연구
추연일,이강현,공준석,이희영,전준호,박종진,김상철,Choo, Yeon Il,Lee, Kang Hyun,Kong, Joon Seok,Lee, Hee Young,Jeon, Joon Ho,Park, Jong Jin,Kim, Sang Chul 한국자동차안전학회 2020 자동차안전학회지 Vol.12 No.2
Modern traffic accidents are a complex occurrence. Various indicators are needed to analyze traffic accidents. Countries that have been investigating traffic accidents for a long time accumulate various data to analyze traffic accidents. The Korean In-Depth Accident Study (KIDAS) database collected damaged vehicles and severity of injury caused by Collision Deformation Classification code (CDC code), Abbreviated Injury Scale (AIS), and Injury Severity Score (ISS). As a result of the investigation, data relating to the injuries of the occupants can be easily obtained, but it was difficult to analyze human severity based on the information of the damaged vehicle. This study suggests a method to measure the speed change at the time of an accident, which is one of the most important indicators in the vehicle crash database, to help advance KIDAS research.