http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
문맥 표현과 셀프 어텐션을 이용한 한국어 영화평 감성 분석
박천음(Cheoneum Park),이동헌(Dongheon Lee),김기훈(Kihoon Kim),이창기(Changki Lee),김현기(Hyunki Kim) 한국정보과학회 2019 정보과학회논문지 Vol.46 No.9
감성 분석은 특정 대상에 대한 의견을 수집하고 분류하는 과정이다. 그러나 자연어에 포함된 사람의 주관을 파악하는 일은 어려운 일로써, 기존의 감성 단어 사전이나 확률 모델은 이러한 문제를 해결하기 어려웠으나 딥 러닝의 발전으로 문제 해결을 시도할 수 있게 됐다. 셀프 어텐션(self-attention)은 주어진 입력열 자신에 대하여 어텐션을 계산하고 가중치 합으로 문맥 벡터를 만들어 모델링하는 방법이며, 문맥상 비슷한 의미를 가진 단어들 간에 높은 가중치가 계산되는 효과가 있다. 본 논문에서는 사전 학습된 문맥 표현을 한국어 감성 분석에 활용하고, 셀프 어텐션으로 모델링하는 방법을 제안한다. 실험 결과, NSMC의 경우 정확도 89.82%, 다음카카오의 경우 92.25%의 성능을 보였다. Sentiment analysis is the processing task that involves collecting and classifying opinions about a specific object. However, it is difficult to grasp the subjectivity of a person using natural language, so the existing sentimental word dictionaries or probabilistic models cannot solve such a task, but the development of deep learning made it possible to solve the task. Self-attention is a method of modeling a given input sequence by calculating the attention weight of the input sequence itself and constructing a context vector with a weighted sum. In the context, a high weight is calculated between words with similar meanings. In this paper, we propose a method using a modeling network with self-attention and pre-trained contextualized embedding to solve the sentiment analysis task. The experimental result shows an accuracy of 89.82%.
U. Yong(용의중),D. Kim(김동환),H. Kim(김호중),D. G. Hwang(황동규),S. Cho(조성건),H. Nam(남효영),S. Kim(김세진),T. Y. Kim(김태영),U. Jeong(정운룡),K. Kim(김기훈),W. K. Chung(정완균),W. H. Yeo(여운홍),J. Jang(장진아) Korean Society for Precision Engineering 2021 한국정밀공학회 학술발표대회 논문집 Vol.2021 No.11월
Over the years, engineered heart tissue (EHT), composed of cardiac cells and a hydrogel, has been considered as a promising in-vitro cardiac model in that it can reproduce the physiological contractions of an actual animal heart. In particular, the contractile force of EHT is one of the representative factors to evaluate drug-induced cardiotoxicity that is a major cause of the withdrawal of drug development. Although there have been a lot of methods to monitor the contractile force of the EHT, most of them are based on optical readout systems that have to process a huge amount of image data. Recently, a strain gauge-based microphysiological system was developed to monitor the contractile force of a laminar cardiac tissue, which can acquire real-time data with a relatively small amount of data. However, the system can monitor only few layers of cardiomyocytes, which are physiologically less relevant environment compared to EHT. Here, we developed a hybrid bioprinted tissue platform, consisting of six bipillar-grafted strain gauges (BPSGs) and one wireless device, that enables online monitoring of the contractile forces from 6 different EHTs in real time during culturing. Furthermore, we confirmed that our system can detect the effects of commercially available drugs on EHTs.