http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
Track 5 : 정보과학영재; 웹앱을 이용한 “광물아! 넌 누구니?”
박예찬 ( Ye Chan Park ),전용주 ( Yong Ju Jeon ) 한국컴퓨터교육학회 2014 한국컴퓨터교육학회 학술발표대회논문집 Vol.18 No.1
이 연구의 목적은 먼저 내가 찾은 광물에 대해서 쉽게 알 수 있고 그 광물을 잘 아는 사람들하고 같이 연구할 수 있게 하는 것이다. 또 전문가와 연구를 같이 하면서로부터 새로운 물질을 발견해내고 그것을 용이하게 사용할 수 있는 것이다. 그리고 우리는 돌은 그냥 땅바닥에 굴러다니는 것이라고만 생각하지만 그 돌이 희귀한 광물일 수도 있기 때문에 우리도 돌에게도 관심을 보이고, 생각해보고, 연구해보도록 할 수 있게 하는 것이다. 본 연구에서는 우선 광물이 무엇인지 그리고 우리 주변에 돌아다니는 돌이 무슨 돌인지 알 수 있게하는 웹앱을 만들고자 한다. 이 웹앱을 통해서 사람들이 돌, 그리고 광물에 대해서 자세히 알고 갈 수 있으면 좋겠다.
측면기반 문장 분석을 기반으로 한 특정대상으로의 보편적 감정 추론
박예찬(Ye-Chan Park),이채린(Chaelyn Lee),이재성(Jaesung Lee) 대한전자공학회 2023 대한전자공학회 학술대회 Vol.2023 No.6
Aspect-Based Sentiment Analysis (ABSA) involves extracting aspects of a target and inferring emotions related to those aspects, and various models have been proposed for this purpose. To do this, a model that can capture the structural characteristics of the target well is needed. In ABSA, LSTM and attention mechanisms are commonly used models, and there have been many studies on models that can capture the structural characteristics of the target. Therefore, in this study, we attempted to train models that combine LSTM and attention mechanisms under common conditions, which are frequently used in ABSA. We also implemented the transcap model, which is a model that does not use attention mechanisms but is good at identifying the structural characteristics of the target, for comparison purposes. The experimental results showed that the SDGCN model had the highest F1 score and that the performance level can vary depending on the way attention mechanisms are used under similar conditions. In addition, the study revealed the performance expectations of attention mechanisms in ABSA classification models, suggesting that various attempts and research are needed to improve their performance.
한국 법률 판결의 자동 분류 및 예측을 위한 트랜스포머 기반 데이터셋 제안
박예찬(Ye-Chan Park),이한용(Hanyong Lee),이재성(Jaesung Lee) 대한전자공학회 2024 대한전자공학회 학술대회 Vol.2024 No.6
This study introduces a novel dataset speifically designed to address the unique challenges of the Korean legal system and to capture the nuanced complexities of Korean legal texts. The dataset encompasses a comprehensive collection of decision documents, essential for training deep learning models to accurately interpret the distinct characteristics of legal language and reasoning inherent in Korean judgments. Extensive evaluations of these models demonstrate their enhanced capability to significantly improve prediction accuracy of legal outcomes. By focusing on the development of this dataset, the research contributes substantially to the field of legal Natural Language Processing (NLP). It not only advances predictive tools tailored for Korean legal texts practitioners looking to streamline decision-making processes. The implications of this dataset extend beyond national borders, suggesting a promising avenue for future research into the global integration of AI within legal systems.
법원의 판단과정의 실증적 검증을 위한 폭행과 협박의 정도에 대한 한국 대법원의 판단기준 분석
박예찬 ( Park Ye-chan ),이재성 ( Lee Jaesung ) 중앙대학교 인문콘텐츠연구소 2022 인공지능인문학연구 Vol.10 No.-
The degree of assault and intimidation is difficult for the general public to analyze because it is unknown how the degree is derived from acting as a factor in dealing with the severity of the crime. Therefore, many studies have been conducted to prove or discover what key elements constitute abstract concepts empirically. This study presents a judgment analysis method that derives elements for each issue using the standard form of the judgment shared through precedents. This is different from previous studies because it allows data mining on large amounts of judgments. This method determines the existence based on the criteria presented by the judgment and establishes the classification criteria according to the theory, and linear regression results are derived. This linear regression model confirmed that the importance of elements by the issue was different, and the reason for the judgment was proved. This could increase the predictability of judgments by analyzing precedents even for the general public.