• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.281-286
• /
• 2021

뉴스 헤드라인에 제3자의 발언을 직접 인용해 전언하는 이른바 '따옴표 저널리즘'이 언론 보도의 객관주의 원칙을 해치는지는 언론학 및 뉴스 구독자에게 중요한 문제이다. 이 연구는 온라인 포털사이트를 통해 실시간 유통되는 한국어 기사의 정확성을 판별하기 위한 기계학습(Machine Learning) 모델을 제안한다. 이 연구에서 제안하는 모델은 Edit Distance와 FastText 기법을 활용해 기사 제목과 본문 내 인용구의 유사성을 측정하고, XGBoost 모델을 활용해 최종 분류한다. 아울러 이 모델을 통해 229만 건의 뉴스 헤드라인에 대해 직접 인용구가 포함된 기사가 취재원의 발언을 주관적인 윤색없이 독자들에게 전하고 있는지를 판별했다. 이뿐만 아니라 딥러닝 기반의 KoELECTRA 모델을 활용해 기사의 제목 내 인용구에 대한 감성 분석을 진행했다. 분석 결과, 윤색이 가미되지 않은 직접 인용형 기사의 비율이 지난 20년 동안 10% 이상 증가했으며, 기사 제목의 인용구에 나타나는 감정은 부정 감성이 긍정 감성의 2.8배 정도로 우세했다. 이러한 시도는 앞으로 계산사회과학 방법론과 빅데이터에 기반한 언론 보도의 평가 및 개선에 도움을 주리라 기대한다.

• Journal of Korea Multimedia Society
• /
• v.17 no.2
• /
• pp.232-239
• /
• 2014

As online texts have been rapidly growing, their automatic classification gains more interest with machine learning methods. Nevertheless, comparatively few research could be found, aiming for Korean texts. Evaluating them with statistical methods are also rare. This study took a sample of tweets and used machine learning methods to classify emotions with features of morphemes and n-grams. As a result, about 76% of emotions contained in tweets was correctly classified. Of the two methods compared in this study, Support Vector Machines were found more accurate than Na$\ddot{i}$ve Bayes. The linear model of SVM was not inferior to the non-linear one. Morphological features did not contribute to accuracy more than did the n-grams.

• Journal of the Korea Convergence Society
• /
• v.8 no.1
• /
• pp.25-34
• /
• 2017

With the recent spread of smartphones and the introduction of web services, online users can access large-scale content regardless of time or place. However, users have had trouble finding the content they wanted among large-scale content. To solve this problem, user modeling and content recommendation system have been actively studied in various fields. However, in spite of active changes in senior groups according to the changes in information environment, research on user modeling and content recommendation system focused on senior groups are insufficient. In this paper, we propose a method of modeling smart senior based on their preference, and further develop a smart senior classification model using machine learning methods. As a result, we can not only grasp the preferences of smart seniors, but also develop a smart senior classification model, which is the foundation for the research of a recommendation system which will provide the activities and contents most suitable for senior groups.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2020.11a
• /
• pp.975-977
• /
• 2020

감정 분석은 문서의 주관적인 감정, 의견, 기분을 파악하기 위한 방법으로 소셜 미디어, 온라인 리뷰 등 다양한 분야에서 활용된다. 문서 내 텍스트가 나타내는 단어와 문맥을 기반으로 감정 수치를 계산하여 긍정 또는 부정 감정을 결정한다. 2015년에 구축된 네이버 영화평 데이터 20 만개에 12 만개를 추가 구축하여 감정 분석 연구를 진행하였으며 언어 모델로는 최근 자연어처리 분야에서 높은 성능을 보여주는 BERT 모델을 이용하였다. 감정 분석 기법으로는 LSTM(Long Short-Term Memory) 등 기존의 기계학습 기법과 구글의 다국어 BERT 모델, 그리고 KoBERT 모델을 이용하여 감정 분석의 성능을 비교하였으며, KoBERT 모델이 89.90%로 가장 높은 성능을 보여주었다.

• KNOM Review
• /
• v.22 no.1
• /
• pp.68-76
• /
• 2019

Bitcoin, based on the blockchain technology is an online crypto-currency developed by Satoshi Nagamoto. Bitcoin, which was first issued on January 3, 2009, is rapidly evolving with increasing number of transactions. However, untoward incidents are occurring due to an increase in the number of Bitcoin transactions. Predicting the number of Bitcoin transactions is important to prepare for any issues that can occur in the Bitcoin network. This paper proposes to design model for predicting the number of Bitcoin transactions by applying two machine learning algorithms and then a model for predicting the number of Bitcoin transactions through experiments.

• Journal of the Architectural Institute of Korea Structure & Construction
• /
• v.34 no.6
• /
• pp.63-69
• /
• 2018

The machine learning model can capture the dynamics of building systems with less inputs than the first principle based simulation model. The training data for developing a machine learning model are usually selected in a heuristic manner. In this study, the authors developed a machine learning model which can describe supply air temperature from an AHU in a real office building. For rational reduction of the training data, the progressive sampling method was used. It is found that even though the progressive sampling requires far less training data (n=60) than the offline regular sampling (n=1,799), the MBEs of both models are similar (2.6% vs. 5.4%). In addition, for the update of the machine learning model, the normalized mutual information (NMI) was applied. If the NMI between the simulation output and the measured data is less than 0.2, the model has to be updated. By the use of the NMI, the model can perform better prediction ($5.4%{\rightarrow}1.3%$).

• The Journal of the Korea Contents Association
• /
• v.17 no.1
• /
• pp.39-49
• /
• 2017

As the Internet technology advances, data on the web is increasing sharply. Many research study about incremental learning for classifying effectively in data increasing. Web document contains the time-series data such as published date. If we reflect time-series data to classification, it will be an effective classification. In this study, we analyze the time-series variation of the words. We propose an efficient classification through dividing the dataset based on the analysis of time-series information. For experiment, we corrected 1 million online news articles including time-series information. We divide the dataset and classify the dataset using SVM and $Na{\ddot{i}}ve$ Bayes. In each model, we show that classification performance is increasing. Through this study, we showed that reflecting time-series information can improve the classification performance.

• Journal of the Korea Institute of Information Security & Cryptology
• /
• v.27 no.6
• /
• pp.1419-1429
• /
• 2017

Recently, environment based authentication technique had proposed reinforced authentication, which generating statistical model per user after user login history classifies into account takeover or legitimate login. But reinforced authentication is likely to be attacked if user was not attacked in past. To improve this problem in this paper, we propose unconsciousness authentication technique that generates 2-Class user model, which trains user's environmental information and others' one using machine learning algorithms. To evaluate performance of proposed technique, we performed evasion attacks: non-knowledge attacker that does not know any information about user, and sophisticated attacker that only knows one information about user. Experimental results against non-knowledge attacker show that precision and recall of Class 0 were measured as 1.0 and 0.998 respectively, and experimental results against sophisticated attacker show that precision and recall of Class 0 were measured as 0.948 and 0.998 respectively.

• Journal of KIISE
• /
• v.43 no.8
• /
• pp.927-932
• /
• 2016

With the explosive increase in the number of online medical documents, the demand for question-answering systems is increasing. Recently, question-answering models based on machine learning have shown high performances in various domains. However, many question-answering models within the medical domain are still based on information retrieval techniques because of sparseness of training data. Based on various information retrieval techniques, we propose an answer snippet retrieval model for question-answering systems of medical documents. The proposed model first searches candidate answer sentences from medical documents using a cluster-based retrieval technique. Then, it generates reliable answer snippets using a re-ranking model of the candidate answer sentences based on various sentence retrieval techniques. In the experiments with BioASQ 4b, the proposed model showed better performances (MAP of 0.0604) than the previous models.

• Annual Conference on Human and Language Technology
• /
• 2020.10a
• /
• pp.304-309
• /
• 2020

본 연구에서는 한국어 전자사전 DECO(Dictionnaire Electronique du COreen)와 다단어(Multi-Word Expressions: MWE) 개체명을 부분 패턴으로 기술하는 부분문법그래프(Local-Grammar Graph: LGG) 프레임에 기반하여 반자동으로 개체명주석 코퍼스 DecoNAC을 구축한 후, 이를 개체명 분석에 활용하고 또한 기계학습에 필요한 도메인별 학습 데이터로 활용하는 DecoNERO 개체명인식 플랫폼을 소개하는 데에 목적을 두었다. 최근 들어 좋은 성과를 보이는 것으로 보고되고 있는 기계학습 방법론들은 다양한 도메인을 기반으로한 대규모의 학습데이터를 필요로 한다. 본 연구에서는 정교하게 설계된 개체명 사전과 다단어 개체명 시퀀스에 대한 언어자원을 바탕으로 하는 반자동으로 학습데이터를 생성하는 방법론을 제안하였다. 본 연구에서 제안된 개체명주석 코퍼스 DecoNAC 기반 접근법의 성능을 실험하기 위해 온라인 뉴스 기사 텍스트를 바탕으로 실험을 진행하였다. 이 실험에서 DecoNAC을 적용한 경우, KoBERT 모델만으로 개체명을 인식한 결과에 비해 약 7.49%의 성능향상을 기대할 수 있음을 확인하였다.