http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
워드 임베딩과 단어 네트워크 분석을 활용한 비지도학습 기반의 문서 다중 범주 가중치 산출
정재윤(Jaeyun Jeong),모경현(Kyoung Hyun Mo),서승완(Seungwan Seo),김창엽(Czang Yeob Kim),김해동(Haedong Kim),강필성(Pilsung Kang) 대한산업공학회 2018 대한산업공학회지 Vol.44 No.6
Due to the increased amounts of online documents, there is a growing demand for text categorization that categorizes documents into predefined categories. Many approaches to this problem are based on supervised machine learning which couldn’t be applied to unlabeled data. However, large number of documents, such as online cell phone reviews, have no category information and key categories are not predefined. To solve these problems, we propose unsupervised document multi-labeling method based on word embedding and word network analysis. After embedding words in a lower dimensional space using Word2Vec technique, we generate a weight matrix by calculating similarities between words. We create a word network using this matrix and extract the key categories from this network. With key category-weight matrix and co-occurrence matrix, we generate a document-category score matrix. To verify our proposed method, we collect 298,206 cell phone reviews from four review websites. Then, we compared the results of the proposed method with labeled documents from human cognitive perspective.
정민성(Minsung Jeong),최희정(Heejeong Choi),서승완(Seungwan Seo),손규빈(Gyubin Son),박경찬(Kyoungchan Park),강필성(Pilsung Kang) 대한산업공학회 2020 대한산업공학회지 Vol.46 No.2
Modeling sentence similarity plays an important role in natural language processing tasks such as question answering and plagiarism detection. Measuring semantic relationship of two sentences is challenging because of the variability and ambiguity of linguistic expression. Previous studies on sentence similarity are focusing on the configuration of input data and classification model structure. However, we focus on the sentence understanding process of human. Human brain stimulates association effect when one tries to understand a sentence describing landscape or object. The association effect that transforms text into image makes human robust to expression changes and word order changes in a sentence. To implement the association effect, we propose a new sentence similarity model based on Siamese network and Text2image generative adversarial network (GAN). The role of Siamese network is to compute the similarity between two sentences with the shared network weights. Inside the Siamese network, two subnetworks are composed of Text2image GAN which transforms the text data into image data. Once the sentences are transformed into image, latent features are extracted through VGGNet. The sentence similarity is computed from the normalized distance between two feature vectors. To verify our proposed method, we modify the MSCOCO dataset and experimental results show that the proposed method outperforms the benchmarked models without association process.
Convolutional LSTM을 이용한 유의 파고 및 파향의 실시간 추정 기법 연구
노영빈(Youngbin Ro),최희정(Heejeong Choi),이정호(Jungho Lee),서승완(Seungwan Seo),강필성(Pilsung Kang) 대한산업공학회 2020 대한산업공학회지 Vol.46 No.6
Real-time estimation of wave condition is essential to improve sailing efficiency. However, existing methodologies are uneconomical due to the expensive radar and high computational complexity. To this end, we propose a neural network model capable of real-time estimation of significant wave height and direction by using raw ocean images collected from operating vessels. In the proposed method, multiple consecutive ocean images are concatenated as a single clip. Then, Convolutional Long Short-Term Memory (ConvLSTM), which combines Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM), was trained on the clips. The final estimation is performed through regression or classification using the extracted spatiotemporal feature map. Based on the datasets collected from two different ships, our proposed method achieved the absolute error of 8cm and a relative error of 5% for significant wave height estimation. Besides, the proposed method yielded an absolute error of 6° for wave direction.