• Journal of the Korean Society for information Management
• /
• v.39 no.1
• /
• pp.91-117
• /
• 2022

The number of depressed patients in Korea and around the world is rapidly increasing every year. However, most of the mentally ill patients are not aware that they are suffering from the disease, so adequate treatment is not being performed. If depressive symptoms are neglected, it can lead to suicide, anxiety, and other psychological problems. Therefore, early detection and treatment of depression are very important in improving mental health. To improve this problem, this study presented a deep learning-based depression tendency model using Korean social media text. After collecting data from Naver KonwledgeiN, Naver Blog, Hidoc, and Twitter, DSM-5 major depressive disorder diagnosis criteria were used to classify and annotate classes according to the number of depressive symptoms. Afterwards, TF-IDF analysis and simultaneous word analysis were performed to examine the characteristics of each class of the corpus constructed. In addition, word embedding, dictionary-based sentiment analysis, and LDA topic modeling were performed to generate a depression tendency classification model using various text features. Through this, the embedded text, sentiment score, and topic number for each document were calculated and used as text features. As a result, it was confirmed that the highest accuracy rate of 83.28% was achieved when the depression tendency was classified based on the KorBERT algorithm by combining both the emotional score and the topic of the document with the embedded text. This study establishes a classification model for Korean depression trends with improved performance using various text features, and detects potential depressive patients early among Korean online community users, enabling rapid treatment and prevention, thereby enabling the mental health of Korean society. It is significant in that it can help in promotion.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.461-464
• /
• 2022

법원도서관이 발간하는 판례공보를 기반으로 판결문 자동요약을 위한 학습 데이터들이 구축되고 있다. 그런데 판결문 요약에서는 뉴스 요약과는 달리 추출요약과 생성요약 방식이 함께 사용되는 특수성이 있고, 이러한 특수성 때문에 현재 판결문 요약 데이터셋이 요약 프로그램의 성능 향상을 이끌지 못하고 있다고 생각된다. 따라서 법률가들이 판결문을 요약하는 방식을 반영하여, 추출요약 방식으로 작성된 판결요지와 생성요약 방식으로 작성된 판결요지를 분리해서 요약 데이터셋을 만들 필요가 있다. 추출요약과 생성요약에 관한 데이터셋을 따로 구축하기 위해서는 판례공보의 판결요지를 추출요약과 생성요약으로 분류하는 작업이 필요한데, 감성 분석에 사용되는 알고리즘이 판결요지의 분류 작업에 응용될 수 있다는 것을 실험 결과로 알 수 있었다.

• Science of Emotion and Sensibility
• /
• v.15 no.1
• /
• pp.97-104
• /
• 2012

Recently, a big text data has been produced by users, an opinion mining to analyze information and opinion about users is becoming a hot issue. Of the opinion mining, especially a sentiment analysis is a study for analysing emotions such as a positive, negative, happiness, sadness, and so on analysing personal opinions or emotions for commercial products, social issues and opinions of politician. To analyze the sentiment analysis, previous studies used a mapping method setting up a distribution of emotions using two dimensions composed of a valence and arousal. But previous studies set up a distribution of emotions arbitrarily. In order to solve the problem, we composed a distribution of 12 emotions through carrying out a survey using Korean emotion words list. Also, certain emotional states on two dimension overlapping multiple emotions, we proposed a selection method with Roulette wheel method using a selection probability. The proposed method shows to classify a text into emotion extracting emotion terms from a text.

• Annual Conference on Human and Language Technology
• /
• 2016.10a
• /
• pp.195-200
• /
• 2016

본 연구는 MUSE 감성 코퍼스를 활용하여 문장의 극성과 키워드의 극성이 얼마만큼 일치하고 일치하지 않은지를 분석함으로써 특히 문장의 극성과 키워드의 극성이 불일치하는 유형에 대한 연구의 필요성을 역설하고자 한다. 본 연구를 위하여 DICORA에서 구축한 MUSE 감성주석코퍼스 가운데 IT 리뷰글 도메인으로부터 긍정 1,257문장, 부정 1,935문장을, 맛집 리뷰글 도메인으로부터는 긍정 2,418문장, 부정 432문장을 추출하였다. UNITEX를 이용하여 LGG를 구축한 후 이를 위의 코퍼스에 적용하여 나타난 양상을 살펴 본 결과, 긍 부정 문장에서 반대 극성의 키워드가 실현된 경우는 두 도메인에서 약 4~16%의 비율로 나타났으며, 단일 키워드가 아닌 구나 문장 차원으로 극성이 표현된 경우는 두 도메인에서 약 25~40%의 비교적 높은 비율로 나타났음을 확인하였다. 이를 통해 키워드의 극성에 의존하기 보다는 문장과 키워드의 극성이 일치하지 않는 경우들, 가령 문장 전체의 극성을 전환시키는 극성전환장치(PSD)가 실현된 유형이나 문장 내 극성 어휘가 존재하지 않지만 구 또는 문장 차원의 극성이 표현되는 유형들에 대한 유의미한 연구가 수행되어야 비로소 신뢰할만한 오피니언 자동 분류 시스템의 구현이 가능하다는 것을 알 수 있다.

• Journal of the Korea Society of Computer and Information
• /
• v.25 no.6
• /
• pp.81-89
• /
• 2020

Online reviews we encounter commonly on SNS, although a complex range of assessment information affecting the consumer's preferences are included, it is general that such information is just provided by simple numbers or star ratings. Based on those review types, it is not easy to get specific information that consumers want and use it to make a decision for purchase. Therefore, in this study, we propose a prediction methodology that can provide ratings broken down by evaluation items by performing sentiment analysis on restaurant reviews written in Korean. To this end, we select 'food', 'price', 'service', and 'atmosphere' as the main evaluation items of restaurants, and build a new sentiment dictionary for each evaluation item. It also classifies review sentences by rating item, predicts granular ratings through sentiment analysis, and provides additional information that consumers can use to make decisions. Finally, using MAE and RMSE as evaluation indicators it shows that the rating prediction accuracy of the proposed methodology has been improved than previous studies and presents the use case of proposed methodology.

• Journal of Intelligence and Information Systems
• /
• v.21 no.2
• /
• pp.49-67
• /
• 2015

Recently, emerging the notion of big data and social media has led us to enter data's big bang. Social networking services are widely used by people around the world, and they have become a part of major communication tools for all ages. Over the last decade, as online social networking sites become increasingly popular, companies tend to focus on advanced social media analysis for their marketing strategies. In addition to social media analysis, companies are mainly concerned about propagating of negative opinions on social networking sites such as Facebook and Twitter, as well as e-commerce sites. The effect of online word of mouth (WOM) such as product rating, product review, and product recommendations is very influential, and negative opinions have significant impact on product sales. This trend has increased researchers' attention to a natural language processing, such as a sentiment analysis. A sentiment analysis, also refers to as an opinion mining, is a process of identifying the polarity of subjective information and has been applied to various research and practical fields. However, there are obstacles lies when Korean language (Hangul) is used in a natural language processing because it is an agglutinative language with rich morphology pose problems. Therefore, there is a lack of Korean natural language processing resources such as a sentiment lexicon, and this has resulted in significant limitations for researchers and practitioners who are considering sentiment analysis. Our study builds a Korean sentiment lexicon with collective intelligence, and provides API (Application Programming Interface) service to open and share a sentiment lexicon data with the public (www.openhangul.com). For the pre-processing, we have created a Korean lexicon database with over 517,178 words and classified them into sentiment and non-sentiment words. In order to classify them, we first identified stop words which often quite likely to play a negative role in sentiment analysis and excluded them from our sentiment scoring. In general, sentiment words are nouns, adjectives, verbs, adverbs as they have sentimental expressions such as positive, neutral, and negative. On the other hands, non-sentiment words are interjection, determiner, numeral, postposition, etc. as they generally have no sentimental expressions. To build a reliable sentiment lexicon, we have adopted a concept of collective intelligence as a model for crowdsourcing. In addition, a concept of folksonomy has been implemented in the process of taxonomy to help collective intelligence. In order to make up for an inherent weakness of folksonomy, we have adopted a majority rule by building a voting system. Participants, as voters were offered three voting options to choose from positivity, negativity, and neutrality, and the voting have been conducted on one of the largest social networking sites for college students in Korea. More than 35,000 votes have been made by college students in Korea, and we keep this voting system open by maintaining the project as a perpetual study. Besides, any change in the sentiment score of words can be an important observation because it enables us to keep track of temporal changes in Korean language as a natural language. Lastly, our study offers a RESTful, JSON based API service through a web platform to make easier support for users such as researchers, companies, and developers. Finally, our study makes important contributions to both research and practice. In terms of research, our Korean sentiment lexicon plays an important role as a resource for Korean natural language processing. In terms of practice, practitioners such as managers and marketers can implement sentiment analysis effectively by using Korean sentiment lexicon we built. Moreover, our study sheds new light on the value of folksonomy by combining collective intelligence, and we also expect to give a new direction and a new start to the development of Korean natural language processing.

• Annual Conference on Human and Language Technology
• /
• 2013.10a
• /
• pp.113-115
• /
• 2013

다수의 상품평 요약은 인터넷 쇼핑몰 고객에게 편의를 제공할 수 있다. 본 논문에서는 상품평 요약 시스템의 성능 향상을 위한 방안을 제안한다. 시스템은 크게 상품평의 평가 항목 추출과 극성 사전 생성, 극성 판별 단계로 구성된다. 평가 항목 추출에서는 외부 연관도의 영향력을 줄이고, 극성 사전 생성에서는 단어 거리 평균을 적용한다. 제안한 방식을 사용하였을 때 평가 항목에 대한 문장의 극성 판별 시 90.8%의 정확율을 보였다.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.320-325
• /
• 2021

BERT는 사전 학습 단계에서 다음 문장 예측 문제와 마스킹된 단어에 대한 예측 문제를 학습하여 여러 자연어 다운스트림 태스크에서 높은 성능을 보였다. 본 연구에서는 BERT의 사전 학습 문제 중 다음 문장 예측 문제에 대해 주목했다. 다음 문장 예측 문제는 자연어 추론 문제와 질의 응답 문제와 같이 임의의 두 문장 사이의 관계를 모델링하는 문제들에 성능 향상을 위해 사용되었다. 하지만 BERT의 다음 문장 예측 문제는 두 문장을 특수 토큰으로 분리하여 단일 문자열 형태로 모델에 입력으로 주어지는 cross-encoding 방식만을 학습하기 때문에 문장을 각각 인코딩하는 bi-encoding 방식의 다운스트림 태스크를 고려하지 않은 점에서 아쉬움이 있다. 본 논문에서는 기존 BERT의 다음 문장 예측 문제를 확장하여 bi-encoding 방식의 다음 문장 예측 문제를 추가적으로 사전 학습하여 단일 문장 분류 문제와 문장 임베딩을 활용하는 문제에서 성능을 향상 시키는 Bi-Cross 사전 학습 기법을 소개한다. Bi-Cross 학습 기법은 영화 리뷰 감성 분류 데이터 셋인 NSMC 데이터 셋에 대해 학습 데이터의 0.1%만 사용하는 학습 환경에서 Bi-Cross 사전 학습 기법 적용 전 모델 대비 5점 가량의 성능 향상이 있었다. 또한 KorSTS의 bi-encoding 방식의 문장 임베딩 성능 평가에서 Bi-Cross 사전 학습 기법 적용 전 모델 대비 1.5점의 성능 향상을 보였다.

• Journal of Intelligence and Information Systems
• /
• v.24 no.2
• /
• pp.59-83
• /
• 2018

With the increasing importance of sentiment analysis to grasp the needs of customers and the public, various types of deep learning models have been actively applied to English texts. In the sentiment analysis of English texts by deep learning, natural language sentences included in training and test datasets are usually converted into sequences of word vectors before being entered into the deep learning models. In this case, word vectors generally refer to vector representations of words obtained through splitting a sentence by space characters. There are several ways to derive word vectors, one of which is Word2Vec used for producing the 300 dimensional Google word vectors from about 100 billion words of Google News data. They have been widely used in the studies of sentiment analysis of reviews from various fields such as restaurants, movies, laptops, cameras, etc. Unlike English, morpheme plays an essential role in sentiment analysis and sentence structure analysis in Korean, which is a typical agglutinative language with developed postpositions and endings. A morpheme can be defined as the smallest meaningful unit of a language, and a word consists of one or more morphemes. For example, for a word '예쁘고', the morphemes are '예쁘(= adjective)' and '고(=connective ending)'. Reflecting the significance of Korean morphemes, it seems reasonable to adopt the morphemes as a basic unit in Korean sentiment analysis. Therefore, in this study, we use 'morpheme vector' as an input to a deep learning model rather than 'word vector' which is mainly used in English text. The morpheme vector refers to a vector representation for the morpheme and can be derived by applying an existent word vector derivation mechanism to the sentences divided into constituent morphemes. By the way, here come some questions as follows. What is the desirable range of POS(Part-Of-Speech) tags when deriving morpheme vectors for improving the classification accuracy of a deep learning model? Is it proper to apply a typical word vector model which primarily relies on the form of words to Korean with a high homonym ratio? Will the text preprocessing such as correcting spelling or spacing errors affect the classification accuracy, especially when drawing morpheme vectors from Korean product reviews with a lot of grammatical mistakes and variations? We seek to find empirical answers to these fundamental issues, which may be encountered first when applying various deep learning models to Korean texts. As a starting point, we summarized these issues as three central research questions as follows. First, which is better effective, to use morpheme vectors from grammatically correct texts of other domain than the analysis target, or to use morpheme vectors from considerably ungrammatical texts of the same domain, as the initial input of a deep learning model? Second, what is an appropriate morpheme vector derivation method for Korean regarding the range of POS tags, homonym, text preprocessing, minimum frequency? Third, can we get a satisfactory level of classification accuracy when applying deep learning to Korean sentiment analysis? As an approach to these research questions, we generate various types of morpheme vectors reflecting the research questions and then compare the classification accuracy through a non-static CNN(Convolutional Neural Network) model taking in the morpheme vectors. As for training and test datasets, Naver Shopping's 17,260 cosmetics product reviews are used. To derive morpheme vectors, we use data from the same domain as the target one and data from other domain; Naver shopping's about 2 million cosmetics product reviews and 520,000 Naver News data arguably corresponding to Google's News data. The six primary sets of morpheme vectors constructed in this study differ in terms of the following three criteria. First, they come from two types of data source; Naver news of high grammatical correctness and Naver shopping's cosmetics product reviews of low grammatical correctness. Second, they are distinguished in the degree of data preprocessing, namely, only splitting sentences or up to additional spelling and spacing corrections after sentence separation. Third, they vary concerning the form of input fed into a word vector model; whether the morphemes themselves are entered into a word vector model or with their POS tags attached. The morpheme vectors further vary depending on the consideration range of POS tags, the minimum frequency of morphemes included, and the random initialization range. All morpheme vectors are derived through CBOW(Continuous Bag-Of-Words) model with the context window 5 and the vector dimension 300. It seems that utilizing the same domain text even with a lower degree of grammatical correctness, performing spelling and spacing corrections as well as sentence splitting, and incorporating morphemes of any POS tags including incomprehensible category lead to the better classification accuracy. The POS tag attachment, which is devised for the high proportion of homonyms in Korean, and the minimum frequency standard for the morpheme to be included seem not to have any definite influence on the classification accuracy.

• Journal of Intelligence and Information Systems
• /
• v.24 no.1
• /
• pp.227-252
• /
• 2018

The evolution of instant communication has mirrored the development of the Internet and messenger applications are among the most representative manifestations of instant communication technologies. In messenger applications, senders use emoticons to supplement the emotions conveyed in the text of their messages. The fact that communication via messenger applications is not face-to-face makes it difficult for senders to communicate their emotions to message recipients. Emoticons have long been used as symbols that indicate the moods of speakers. However, at present, emoticon-use is evolving into a means of conveying the psychological states of consumers who want to express individual characteristics and personality quirks while communicating their emotions to others. The fact that companies like KakaoTalk, Line, Apple, etc. have begun conducting emoticon business and sales of related content are expected to gradually increase testifies to the significance of this phenomenon. Nevertheless, despite the development of emoticons themselves and the growth of the emoticon market, no suitable emoticon recommendation system has yet been developed. Even KakaoTalk, a messenger application that commands more than 90% of domestic market share in South Korea, just grouped in to popularity, most recent, or brief category. This means consumers face the inconvenience of constantly scrolling around to locate the emoticons they want. The creation of an emoticon recommendation system would improve consumer convenience and satisfaction and increase the sales revenue of companies the sell emoticons. To recommend appropriate emoticons, it is necessary to quantify the emotions that the consumer sees and emotions. Such quantification will enable us to analyze the characteristics and emotions felt by consumers who used similar emoticons, which, in turn, will facilitate our emoticon recommendations for consumers. One way to quantify emoticons use is metadata-ization. Metadata-ization is a means of structuring or organizing unstructured and semi-structured data to extract meaning. By structuring unstructured emoticon data through metadata-ization, we can easily classify emoticons based on the emotions consumers want to express. To determine emoticons' precise emotions, we had to consider sub-detail expressions-not only the seven common emotional adjectives but also the metaphorical expressions that appear only in South Korean proved by previous studies related to emotion focusing on the emoticon's characteristics. We therefore collected the sub-detail expressions of emotion based on the "Shape", "Color" and "Adumbration". Moreover, to design a highly accurate recommendation system, we considered both emotion-technical indexes and emoticon-emotional indexes. We then identified 14 features of emoticon-technical indexes and selected 36 emotional adjectives. The 36 emotional adjectives consisted of contrasting adjectives, which we reduced to 18, and we measured the 18 emotional adjectives using 40 emoticon sets randomly selected from the top-ranked emoticons in the KakaoTalk shop. We surveyed 277 consumers in their mid-twenties who had experience purchasing emoticons; we recruited them online and asked them to evaluate five different emoticon sets. After data acquisition, we conducted a factor analysis of emoticon-emotional factors. We extracted four factors that we named "Comic", Softness", "Modernity" and "Transparency". We analyzed both the relationship between indexes and consumer attitude and the relationship between emoticon-technical indexes and emoticon-emotional factors. Through this process, we confirmed that the emoticon-technical indexes did not directly affect consumer attitudes but had a mediating effect on consumer attitudes through emoticon-emotional factors. The results of the analysis revealed the mechanism consumers use to evaluate emoticons; the results also showed that consumers' emoticon-technical indexes affected emoticon-emotional factors and that the emoticon-emotional factors affected consumer satisfaction. We therefore designed the emoticon recommendation system using only four emoticon-emotional factors; we created a recommendation method to calculate the Euclidean distance from each factors' emotion. In an attempt to increase the accuracy of the emoticon recommendation system, we compared the emotional patterns of selected emoticons with the recommended emoticons. The emotional patterns corresponded in principle. We verified the emoticon recommendation system by testing prediction accuracy; the predictions were 81.02% accurate in the first result, 76.64% accurate in the second, and 81.63% accurate in the third. This study developed a methodology that can be used in various fields academically and practically. We expect that the novel emoticon recommendation system we designed will increase emoticon sales for companies who conduct business in this domain and make consumer experiences more convenient. In addition, this study served as an important first step in the development of an intelligent emoticon recommendation system. The emotional factors proposed in this study could be collected in an emotional library that could serve as an emotion index for evaluation when new emoticons are released. Moreover, by combining the accumulated emotional library with company sales data, sales information, and consumer data, companies could develop hybrid recommendation systems that would bolster convenience for consumers and serve as intellectual assets that companies could strategically deploy.