http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
KoBERT에서 데이터 불균형 문제 경감을 위한 정렬 알고리즘을 이용한 학습 데이터 구성
고영서(Young-seo Go),최승호(Seoung-Ho Choi) 한국디지털콘텐츠학회 2023 한국디지털콘텐츠학회논문지 Vol.24 No.7
Models trained with imbalanced data do not perform well in classification for a small sample of abnormal data. We aim to reduce the imbalance problem in natural language processing by using a sorting algorithm. The computation cost of each sorting algorithm is measured and compared to derive an algorithm suitable for data processing. The sorted data are preprocessed according to three imbalance criteria to create training data and then fine-tuned using KoBERT which is a natural language processing model. The performance of data adjustment was evaluated by measuring accuracy, recall, and precision according to the imbalance scale of the training data. We confirmed that the data imbalance problem in natural language processing could be alleviated by applying the sorting algorithm of the proposed method.
객체 검출을 활용한 교육적 목적의 웹 기반 브레드보드 전기회로 분석
박세현(Se Hyeon Park),최승호(Seoung-Ho Choi) 한국디지털콘텐츠학회 2022 한국디지털콘텐츠학회논문지 Vol.23 No.9
In an experiment where students learn electrical and electronic theory and apply it, students have difficulty in connecting circuits. To alleviate the above difficulties, we first propose a circuit analysis service. The proposed method detects an electric device using an object detection model in which electric devices connected to a breadboard are labeled with data. To verify this, we connect to the breadboard circuit and create a custom dataset through photographs. The proposed method is divided into two processes: electric device prediction and electric device position detection. The electric device prediction model was compared using five object detection models, and the Faster R-CNN model had the best prediction performance. The electrical device position detector extracts features from the object detection model through transition learning to predict two coordinates (x1, y1), (x2, y2). A comparison of each model confirmed that the ResNet model has good location detection performance. Through this, it was confirmed that the proposed method alleviates the difficulty of first-time students learning electric and electronic experiments.