Mining User Behaviors in Social Networks and Its Applications

Byung-Won On(온병원) 한국지능정보시스템학회 2013 한국지능정보시스템학회 학술대회논문집 Vol.2013 No.11

국내 휴대폰의 진화패턴 규명을 위한 텍스트 마이닝 방안 제안 및 사례 연구

온병원(Byung-Won On) 한국컴퓨터정보학회 2015 韓國컴퓨터情報學會論文誌 Vol.20 No.2

생물의 진화패턴과 원리는 지난 200년간 학문적인 영역에서 활발히 연구되어 왔으며 생명의 진화에 대한 체계적인 이론, 개념 및 방법론이 제시되었다. 그리고 진화경제학, 진화심리학, 진화언어학 등 다양한 분야에 적용되어 큰 연구 성과를 거두고 있다. 이와 더불어 진화생물학 논리를 인간이 만든 제품에 적용하려는 시도도 병행되어 왔다. 기존 연구들이 생물진화 논리를 인공물에 그대로 적용하거나 해당 분야 전문가의 직관에 근거하여 진화 모형을 구축하는 것이어서 진화 모형에 대한 일반화를 시키기에는 한계를 가진다. 또한 생물과 달리 인공물은 인간 의지의 상상력이 반영되기 때문에 생물진화 이론을 곧바로 적용할 수 없다고 알려져 왔다. 따라서 본 논문에서는 특정인의 주관에 벗어나 일반 대중들의 생각을 엿보고 이를 바탕으로 진화 모형을 구축하는 것을 목표로 한다. 이를 위해, 인공물을 계통적으로 분류할 수 있는 체계적인 틀을 제시하는 텍스트 마이닝 방안과 그 결과물을 효과적으로 보여줄 수 있는 시각화 방안을 차례로 제안한다. 특히, 제안방안을 바탕으로 최근 혁신의 아이콘으로 떠오르고 있는 휴대폰과 스마트폰에 대한 사례 연구를 집중적으로 수행한다. 지난 10년간 국내에서 출시된 휴대폰과 스마트폰에 대한 리뷰 포스트들을 수집하고 분석하여, 진화패턴을 발견하고 요약해서 보여주며 그 결과에 대해서 자세히 토의한다. 더욱이 이러한 작업은 소수의 전문가들이 방대한 문헌과 자료를 조사 정리하여, 오랜 시간에 걸쳐 진화계통도를 그리게 되는 매우 지난한 작업이다. 하지만 본 논문에서 제안한 방안은 반자동(semi-automatic) 마이닝 알고리즘으로 인간의 노력을 최소화할 수 있어 그 효용 가치가 높다. 이러한 연구를 통해 인간의 창의력과 상상력이 구현되는 방식을 이해하고 휴대폰의 미래 모습을 전망하는데 있어 유관기업들에게 큰 도움을 줄 것이다. Systematic theory, concepts, and methodology for the biological evolution have been developed while patterns and principles of the evolution have been actively studied in the past 200 years. Furthermore, they are applied to various fields such as evolutionary economics, evolutionary psychology, evolutionary linguistics, making significant progress in research. In addition, existing studies have applied main biological evolutionary models to artifacts although such methods do not fit to them. These models are also limited to generalize evolutionary patterns of artifacts because they are designed in terms of a subjective point of view of experts who know well about the artifacts. Unlike biological organisms, because artifacts are likely to reflect the imagination of the human will, it is known that the theory of biological evolution cannot be directly applied to artifacts. In this paper, beyond the individual’s subjective, the aim of our research is to present evolutionary patterns of a given artifact based on peeping the idea of the public. For this, we propose a text mining approach that presents a systematic framework that can find out the evolutionary patterns of a given artifact and then visualize effectively. In particular, based on our proposal, we focus mainly on a case study of mobile phone that has emerged as an icon of innovation in recent years. We collect and analyze review posts on mobile phone available in the domestic market over the past decade, and discuss the detailed results about evolutionary patterns of the mobile phone. Moreover, this kind of task is a tedious work over a long period of time because a small number of experts carry out an extensive literature survey and summarize a huge number of materials to finally draw a diagram of evolutionary patterns of the mobile phone. However, in this work, to minimize the human efforts, we present a semi-automatic mining algorithm, and through this research we can understand how human creativity and imagination are implemented. In addition, it is a big help to predict the future trend of mobile phone in business and industries.

Link Structure based Community Detection 알고리즘의 제안과 소셜 네트워크 분석 및 비주얼라이제이션을 위한 사례 연구

온병원(Byung-Won On),이인규(Ingyu Lee),이만재(Manjai Lee) 한국HCI학회 2012 한국HCI학회 학술대회 Vol.2012 No.1

커뮤니티 스트럭처는 소셜 네트워크를 이해하는 주요한 특성 가운데 하나이다. 커뮤니티 디텍션에 대한 최근 연구에 따르면 실제 소셜 네트워크에서 중첩된 커뮤니티들을 쉽게 발견할 수 있으며 특정 노드들은 하나 이상의 커뮤니티에 속한다. 기존의 방법론들이 중첩된 커뮤니티들을 효과적으로 찾지 못하는 동안 우리는 이 논문에서 링크 클러스터링 알고리즘을 이용한 새로운 커뮤니티 디텍션 알고리즘을 제안한다. 그리고 우리의 방법론을 소셜 네트워크에 적용함으로써 발견된 커뮤니티 스트럭처를 기반으로 하여 주어진 네트워크에 대한 분석 및 비주얼라이제이션을 시도한다. 이러한 사례 연구는 우리의 제안 방안이 우수함을 간접적으로 증명한다. Community structure is one of main characteristics in order to understand social networks. Recent studies on the community detection problem have shown the existence of overlapped communities around real social networks in which some nodes belong to multiple communities. While existing methods seldom find such overlapped communities, in this paper, we propose a novel community detection approach based on the link clustering algorithm. Furthermore, by applying our proposal to a real social network, we study the principles of the given social network by means of community structures found.

BERT 모델의 감성 분류 정확도 향상을 위한 감성 단어 마스킹 방안

이원민(Won-Min Lee),온병원(Byung-Won On) 한국정보기술학회 2021 Proceedings of KIIT Conference Vol.2021 No.6

감성 분석(Sentiment Analysis)이란 텍스트의 정보를 추출하고 감성에 따라 분류하는 과정으로 최근에는 딥러닝을 이용한 연구가 활발히 이루어지고 있다. 그중 BERT는 구글에서 공개한 전이 학습을 통해 좋은 성능을 발휘하는 딥러닝 모델로, 감성 분석에서 많이 사용되는 모델이다. 해당 모델은 마스크 언어 모델 태스크를 수행할 때 랜덤 마스킹을 통한 학습 데이터를 생성한다. 이때 감성 단어로 마스킹을 진행하면 감성에 집중된 학습 데이터를 생성할 수 있고, 감성 분석의 정확도를 높일 수 있다. 따라서 감성 단어로 마스킹을 진행해서 학습 데이터를 생성하고 마스크 언어 모델 태스크를 수행하는 감성 분류에 최적화된 모델을 제안한다. 제안 방안으로 감성 분석을 진행 한 결과, 기존의 방식보다 향상된 성능을 보였다. Sentiment Analysis is a process of extracting text information and classifying it according to emotion. Recently, research using deep learning has been actively conducted. Among them, BERT is a deep learning model released by Google that shows good performance through transfer learning, and is a model that is widely used in sentiment analysis. The model generates training data through random masking when performing the mask language model task. At this time, if masking is performed with emotional words, it is possible to generate learning data focused on the emotions, and to increase the accuracy of the emotion analysis. Therefore, we propose a model optimized for emotional classification that generates training data by masking with emotional words and performs the mask language model task. As a result of conducting emotional analysis as a proposed method, it showed improved performance compared to existing methods.

인공지능 챗봇을 위한 감정 문장 생성 시스템

이원민(Won-Min Lee),온병원(Byung-Won On) 한국정보기술학회 2021 Proceedings of KIIT Conference Vol.2021 No.11

인공지능 시장의 성장과 함께 챗봇에 관한 연구가 활발히 진행되고 있다. 최근 챗봇은 사용자에게 필요한 서비스를 제공해주는 기술을 넘어 사람의 감정을 비슷하게 표현하며 적절한 반응을 보이는 수준으로 기술이 개발되고 있다. 하지만 챗봇이 사람과 똑같이 감정을 이해하고 표현하는 것에는 한계가 존재한다. 본 논문에서는 BERT+GPT 파이프라인 모델을 통해 자동으로 생성된 대용량의 감정 학습데이터를 적용하여 인공지능 챗봇시스템을 구현하였고, 프로토타입 시스템을 통해 실제 챗봇이 사용자의 말에 대한 감정을 파악하고 이에 적절한 반응을 보임을 확인했다. With the growth of the artificial intelligence market, research on chatbots is being actively conducted. In recent years, chatbot technology is being developed to a level where it can respond appropriately by expressing human emotions in a similar way, beyond the technology that provides necessary services to users. However, there is a limit to the ability of chatbots to understand and express emotions just like humans. In this paper, an artificial intelligence chatbot system was implemented by applying a large amount of emotion learning data automatically generated through the BERT+GPT pipeline model. It was confirmed that there was a reaction.

감성 및 감정 단어 마스킹 기반 BERT와 GPT 파이프라인 방식을 통한 감정 문장 생성

이원민(Won-Min Lee),온병원(Byung-Won On) 한국정보기술학회 2021 한국정보기술학회논문지 Vol.19 No.9

Recently, due to advances in Artificial Intelligence(AI), there have been active studies on AI Chatbot, which has reached a level where it can similarly express human emotions. If AI Chatbot can accurately understand human emotions and respond accordingly, natural conversations between chatbots and humans are possible. Large-capacity, high-quality training data for emotion analysis is needed to train a deep learning model that automatically generates sentences. In this study, we proposes a new BERT+GPT pipeline method in which an emotion word masking-based BERT is used to automatically generate large-capacity, high-quality training data as the input of GPT that is used to generate response sentences containing human emotions. Our experimental results show that the proposed method improved up to 12% accuracy, compared to the existing BERT model, showing that most response sentences generated by GPT were better in both emotion consistency and natural meaning.

빅데이터 처리를 위한 메타데이터 표준화에 관한 연구

정동원(Dongwon Jeong),이석훈(Sukhoon Lee),정현준(Hyunjun Jung),온병원(Byung-Won On) 한국정보과학회 2015 한국정보과학회 학술발표논문집 Vol.2015 No.12

공공 데이터 분석결과의 게임화를 통한 웹 방문자 성향 분석

허종욱(Jong-Uk Heo),온병원(Byung-Won On),이만재(Manjai Lee) 한국HCI학회 2013 한국HCI학회 학술대회 Vol.2013 No.1

빅 데이터의 등장과 함께 공공기관이 보유한 데이터를 활용하여 국민에게 개선된 서비스를 제공하고자 하는 정책이 탄력을 받고 있다. 일반 국민이 관심을 갖는 국회의원의 본회의 표결 데이터를 기반으로 국회의 의정활동을 파악하고자 하는 폴리티즈라는 웹 사이트를 개발하였다. 웹 사이트의 기능 중 하나는 실제 국회에서 처리된 법안에 대해 웹 사용자에게 찬반을 묻도록 하였으며 자신과 가까운 성향을 갖는 의원을 알 수 있도록 게이미피케이션을 적용하였다. 방문자의 투표결과를 통해 해당 법안에 대한 국민의 생각을 알 수 있음을 발견하였다. (http://www.politiz.org/에서 서비스 중이며, 구글크롬, IE8.0이상의 웹브라우저에서 사용가능함) The utilization of public data based on Big data analytics would contribute to the development of transparent public policy. As one of case studies, we developed the so-called Politiz web site in which we can understand congressional activities using roll call votes in the 18<SUP>th</SUP> Korean National Assembly. In the web site, a user can vote sixteen controversial bills and finally find the best match showing whom the user is close to among all congressmen. In our research, we showed a congressional social network that provides us with the holistic insight mined from the roll call data based on Big data analytics. In particular, we focus on studying how to apply the concept of gamification so as to make users more fun and engaging in the Politiz web site. In addition, our web site provides a handy tool for collecting public opinion about controversial issues occurring in our society.

잠재 토픽 기반의 제품 평판 마이닝

박상민(Sang-Min Park),온병원(Byung-Won On) 한국지능정보시스템학회 2017 지능정보연구 Vol.23 No.2

Data-drive analytics techniques have been recently applied to public surveys. Instead of simply gathering survey results or expert opinions to research the preference for a recently launched product, enterprises need a way to collect and analyze various types of online data and then accurately figure out customer preferences. In the main concept of existing data-based survey methods, the sentiment lexicon for a particular domain is first constructed by domain experts who usually judge the positive, neutral, or negative meanings of the frequently used words from the collected text documents. In order to research the preference for a particular product, the existing approach collects (1) review posts, which are related to the product, from several product review web sites; (2) extracts sentences (or phrases) in the collection after the pre-processing step such as stemming and removal of stop words is performed; (3) classifies the polarity (either positive or negative sense) of each sentence (or phrase) based on the sentiment lexicon; and (4) estimates the positive and negative ratios of the product by dividing the total numbers of the positive and negative sentences (or phrases) by the total number of the sentences (or phrases) in the collection. Furthermore, the existing approach automatically finds important sentences (or phrases) including the positive and negative meaning to/against the product. As a motivated example, given a product like Sonata made by Hyundai Motors, customers often want to see the summary note including what positive points are in the ‘car design’ aspect as well as what negative points are in thesame aspect. They also want to gain more useful information regarding other aspects such as ‘car quality’, ‘car performance’, and ‘car service.’ Such an information will enable customers to make good choice when they attempt to purchase brand-new vehicles. In addition, automobile makers will be able to figure out the preference and positive/negative points for new models on market. In the near future, the weak points of the models will be improved by the sentiment analysis. For this, the existing approach computes the sentiment score of each sentence (or phrase) and then selects top-k sentences (or phrases) with the highest positive and negative scores. However, the existing approach has several shortcomings and is limited to apply to real applications. The main disadvantages of the existing approach is as follows: (1) The main aspects (e.g., car design, quality, performance, and service) to a product (e.g., Hyundai Sonata) are not considered. Through the sentiment analysis without considering aspects, as a result, the summary note including the positive and negative ratios of the product and top-k sentences (or phrases) with the highest sentiment scores in the entire corpus is just reported to customers and car makers. This approach is not enough and main aspects of the target product need to be considered in the sentiment analysis. (2) In general, since the same word has different meanings across different domains, the sentiment lexicon which is proper to each domain needs to be constructed. The efficient way to construct the sentiment lexicon per domain is required because the sentiment lexicon construction is labor intensive and time consuming. To address the above problems, in this article, we propose a novel product reputation mining algorithm that (1) extracts topics hidden in review documents written by customers; (2) mines main aspects based on the extracted topics; (3) measures the positive and negative ratios of the product using the aspects; and (4) presents the digest in which a few important sentences with the positive and negative meanings are listed in each aspect. Unlike the existing approach, using hidden topics makes experts construct the sentimental lexicon easily and quickly. Furthermore, reinforcing topic semantics, we can improve the accuracy of the product reputati

주제 추출을 위한 맵리듀스 기반의 사전확률 최적화 알고리즘

오선영(SeonYeong Oh),온병원(Byung-Won On) Korean Institute of Information Scientists and Eng 2018 정보과학회논문지 Vol.45 No.5

Various topic extraction algorithms have been used to obtain meaningful information from a large number of text documents. Since the topic extraction algorithms work based on the Bayesian probability model, the prior probabilities, α and β, should be given as inputs. Until now, in order to run the topic extraction models, users have to either take advantage of default prior probability values or determine them subjectively. In this study, we propose a MapReduce-based prior probability optimization algorithm that systematically determines the prior probability values in addition to the improvement of performance and accuracy against a large-scale input data. Unlike the previous single thread algorithm, the proposed MapReduce-based algorithm quickly determines the prior probability values that are suitable for the input data. It then extracts topics with high accuracy after the topic extraction algorithm is executed with the chosen prior probability values. Our experimental results showed that the proposed method outperforms the previous method in the aspect of topic coherence and performance.