• The Journal of the Korea Contents Association
• /
• v.11 no.5
• /
• pp.125-137
• /
• 2011

Recently the research on the system using images taken from camera phones as input is actively conducted. This paper proposed a system that shows wine pictures which are similar to the input wine label in order. For the calculation of the similarity of images, the representative color of each cell of the image, the recognized text color, background color and distribution of feature points are used as the features. In order to calculate the difference of the colors, RGB is converted into CIE-Lab and the feature points are extracted by using Harris Corner Detection Algorithm. The weights of representative color of each cell of image, text color and background color are applied. The image similarity is calculated by normalizing the difference of color similarity and distribution of feature points. After calculating the similarity between the input image and the images in the database, the images in Database are shown in the descent order of the similarity so that the effort of users to search for similar wine labels again from the searched result is reduced.

• Journal of the Korean earth science society
• /
• v.33 no.1
• /
• pp.94-109
• /
• 2012

The purpose of this study is to analyse the descriptive characteristics of the label texts related to Earth Science at a science museum and a natural history museum in Korea. The data were collected from Korean National Science Museum and Seodaemun Natural History Museum. The analysis framework was modified according to the Systemic Functional Linguistics. As a result, characteristics of the labels are 1) mostly declarative sentences, 2) appropriate amount of scientific information, and 3) mainly 'facts'. Moreover, all of the text genre are 4) 'logical expositions'. Particularly in Korean National Science Museum, the labels present 5) more scientific words among the entire terminologies and 6) more than half subjects omitted or long nominalized. Those results may imply that the labels can lead one-way communication regarding the culture of science rather than two-way. This study presents the descriptive characteristics of the label texts to make educationally meaningful communication possible by building an open structure between visitors' own culture in everyday life and the culture of science.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.11a
• /
• pp.439-441
• /
• 2023

전통적인 텍스트 분류 방법은 상당량의 라벨링된 데이터와 미리 정의된 클래스가 필요해서 그 적용성과 확장성이 제한된다. 그래서 이런 한계를 극복하기 위해 제로샷 러닝(Zero-shot Learning)이 등장했다. 텍스트 분류 분야에서 제로샷 텍스트 분류는 모델이 대상 클래스의 샘플을 미리 접하지 않고도 인스턴스를 분류할 수 있도록 하는 중요한 주제이다. 이 문제를 해결하기 위해 정책 네트워크를 활용한 심층 강화 학습(DRL) 기반 접근법을 제안한다. 이러한 방법을 통해 모델이 새로운 의미 공간에 효과적으로 적응하면서, 다른 모델들과 비교하여 제로샷 텍스트 분류의 정확도를 향상시킬 수 있었다. XLM-R 과 비교하면 최대 15.9%의 정확도 향상이 나타났다.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.14-18
• /
• 2023

본 연구는 한국어 음성-텍스트 데이터에서 보편적으로 나타난 휴지의 실현 양상을 분석하고, 이를 토대로 데이터셋을 선별해 보편적이고 규격화된 한국어 휴지 예측을 위한 모델을 제안하였다. 이를 위해 전문적인 발성 훈련을 받은 성우 등의 발화가 녹음된 음성-텍스트 데이터셋을 수집하고 MFA와 같은 음소 정렬기를 사용해 휴지를 라벨링하는 등의 전처리를 하고, 다양한 화자의 발화에서 공통적으로 나타난 휴지를 선별해 학습데이터셋을 구축하였다. 구축된 데이터셋을 바탕으로 LLM 중 하나인 KULLM 모델을 미세 조정하고 제안한 모델의 휴지 예측 성능을 평가하였다.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2024.01a
• /
• pp.13-16
• /
• 2024

온라인상에 인간의 감성을 담은 리뷰 데이터가 꾸준히 축적되어왔다. 이 텍스트 데이터를 분석하고 활용하는 일은 마케팅에 있어서 중요한 자산이 될 것이다. 이와 관련된 Aspect-Based Sentiment Analysis(ABSA) 연구는 한글에 있어서는 데이터 부족을 이유로 거의 선행연구가 없는 실정이다. 본 연구에서는 최근 공개된 데이터 셋을 바탕으로 하여 화장품 도메인에 대한 소비자들의 리뷰 텍스트와 사전 라벨링 된 속성, 감성 극성을 기반으로 ABSA를 진행한다. Klue RoBERTa base 모델을 활용하여 데이터를 학습시키고, Python Kiwipiepy 등으로 전처리한 결과를 대시보드로 시각화하여 분석하기 쉬운 환경을 마련하는 방법을 제시한다.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.368-369
• /
• 2023

대화 데이터 기반 광고 추천은 광고 마케팅에서 고객 맞춤형 광고 제공, 마케팅 효과 극대화 등을 위한 중요한 기술로 주목받고 있다. 본 논문에서는 모바일 인스턴스 메신저인 카카오톡 대화창에서 발생한 텍스트 데이터를 기반으로 대화 내용을 분석하여 대화 주제별 적절한 광고 키워드를 제안한다. 이를 위해 주제별 대화 내용을 미용, 식음료, 상거래로 세분하고 KoNLPy 의 Okt 를 이용하여 텍스트 전처리를 수행하고 키워드별로 빈도수를 뽑아 워드 클라우드를 제시한다. 또한, 잠재 디리클레 할당(Latent Dirichlet Allocation, LDA)을 기반으로 대화 주제를 세분화한 뒤 라벨링을 통해 주제별 대화 키워드를 분석한다. 실험 결과, 대화 주제를 온라인 쇼핑, 헤어, 뷰티 관리, 음식으로 나눌 수 있었으며, 토픽별 상위 키워드를 Word2Vec 을 통해 특정 단어와 유사한 키워드를 도출하여 적절한 광고 키워드를 제시할 수 있었다.

• Korean Journal of Cognitive Science
• /
• v.34 no.3
• /
• pp.197-226
• /
• 2023

Generative Text Summarization is one of the Natural Language Processing tasks. It generates a short abbreviated summary while preserving the content of the long text. ROUGE is a widely used lexical-overlap based metric for text summarization models in generative summarization benchmarks. Although it shows very high performance, the studies report that 30% of the generated summary and the text are still inconsistent. This paper proposes a methodology for evaluating the performance of the summary model without using the correct summary. AggreFACT is a human-annotated dataset that classifies the types of errors in neural text summarization models. Among all the test candidates, the two cases, generation summary, and when errors occurred throughout the summary showed the highest correlation results. We observed that the proposed evaluation score showed a high correlation with models finetuned with BART and PEGASUS, which is pretrained with a large-scale Transformer structure.

• Journal of the Korea Convergence Society
• /
• v.9 no.7
• /
• pp.25-31
• /
• 2018

Online news has replaced the traditional newspaper and has brought about a profound transformation in the way we access and share information. News websites have had the ability for users to post comments for quite some time, and some have also begun to crowdsource reactions to news articles. The field of sentiment analysis seeks to computationally model the emotions and reactions experienced when presented with text. In this work, we analyze more than 100,000 news articles over ten categories with five user-generated emotional annotations to determine whether or not these reactions have a mathematical correlation to the news body text and propose a simple sentiment analysis algorithm that requires minimal preprocessing and no machine learning. We show that it is effective even for a morphologically complex language like Korean.

• The Journal of the Institute of Internet, Broadcasting and Communication
• /
• v.22 no.6
• /
• pp.185-190
• /
• 2022

Open data has a lot of economic value. Not only Korea, but many other countries are doing their best to make various policies and efforts to expand and utilize open data. However, although Korea has a large amount of data, the data is not utilized effectively. Thus, attempts to utilize those data should be made in various industries. In particular, in the fashion industry, exchange and refund problems are the most common due to unpredictable consumers. Better feedback is necessary for service providers to solve this problem. We want to solve it by showing improved images of dissatisfactions along with user reviews including consumer needs. In this paper, user reviews are analyzed on online shopping mall websites to identify consumer needs, and product attributes are defined by utilizing the attributes of K-fashion data. The users' request is defined as a dissatisfaction attribute, and labeling data with the corresponding attribute is searched. The users' request is provided to the service provider in forms of text data or attributes, as well as an image to help improve the product.

• Convergence Security Journal
• /
• v.23 no.4
• /
• pp.61-70
• /
• 2023

Clustering is an unsupervised learning method that involves grouping data based on features such as distance metrics, using data without known labels or ground truth values. This method has the advantage of being applicable to various types of data, including images, text, and audio, without the need for labeling. Traditional clustering techniques involve applying dimensionality reduction methods or extracting specific features to perform clustering. However, with the advancement of deep learning models, research on deep clustering techniques using techniques such as autoencoders and generative adversarial networks, which represent input data as latent vectors, has emerged. In this study, we propose a deep clustering technique based on deep learning. In this approach, we use an autoencoder to transform the input data into latent vectors, and then construct a vector space according to the cluster structure and perform k-means clustering. We conducted experiments using the MNIST and Fashion-MNIST datasets in the PyTorch machine learning library as the experimental environment. The model used is a convolutional neural network-based autoencoder model. The experimental results show an accuracy of 89.42% for MNIST and 56.64% for Fashion-MNIST when k is set to 10.