이경원 GUO WEN JUN 2018-08 28017 https://aurora.ajou.ac.kr/handle/2018.oak/14110 학위논문(박사)--아주대학교 일반대학원 :라이프미디어협동과정,2018. 8 시간적 이벤트 데이터(temporal event data)는 다양한 분야에서 점점 더 많은 관심을 받고 있다. 시간적 이벤트 데이터의 이벤트들이 발생한 시간에 따라 이벤트들의 패턴이나 유사성에 대한 비교 및 분석을 통해서 이벤트의 새로운 구성을 식별하고, 유저에게 향후 의사 결정을 할 때도 많은 도움을 줄 수 있다. 기존에 연구들이 시간의 흐름에 따라 변화하는 이벤트들에 집중적으로 연구를 이루어졌지만 최근에 반복을 인해 생긴 종단적인(longitudinal) 시간의 변화에 대한 연구도 많아지고 있다. 본 연구에서는 종단적인 시간성을 가진 시간적 이벤트 데이터에 비교 분석에 대한 최적화된 시각화 시스템을 제안하고자 한다. 또한 시스템의 활용 사례 연구와 사용성 평가를 통해 다음과 같은 효과를 검증하였다. 첫째, ‘BubbleUp’ 시스템의 시각화 알고리즘 설계와 프로토타입을 구현하고, 추가 인터랙션 기능과 사용자 인터페이스를 개발하였다. 둘째, 클러스터링을 통해 데이터의 잠재적인 패턴 도출이 가능하여 데이터에 대한 탐색이 보다 쉬워졌다. 셋째, 사용자가 타깃을 선택하면 시스템에서 유사한 결과와 유사하지 않는 결과에 대해 제시해 줄뿐만 아니라 사용자가 원하는 분포 범위를 자유롭게 설정할 수 있으며, 설정된 범위기준으로 유사한 결과를 유사도 결과 랭킹 리스트와 함께 제공해 준다. 넷째, 머신러닝을 통해 현재의 데이터들을 기반으로, 향후 결과에 대한 예측과 분석이 가능하였다. “BubbleUp”시스템을 통해서 종단적인 시간적 이벤트 데이터의 유사성, 군집화, 예측에 대한 통합적인 분석을 직관적으로 가능하게 제공하였다. I. 서론 ·················································································1 1. 연구의 배경 ············································································· 1 2. 연구의 필요성 ········································································· 5 3. 연구의 목적 ············································································· 6 II. 관련 연구 ···································································8 1. 시간적 이벤트 데이터 시)화 ··········································· 8 1) 시간적 이벤트 데이터의 정의·············································8 2) 시간적 이벤트 데이터 시각화 사례 분석····························10 2. 유사성 연구 ········································································16 1) 유사성 측정··········································································16 2) 유사성 관련 연구 ································································· 18 3) 다차원 척도법 ······································································ 20 3. 데이터 클러스터링 ································································ 23 1) 데이터 클러스터링·······························································23 2) 데이터 클러스터링 관련 연구 ·············································· 26 4. 머신러닝을 통한 데이터의 예측·········································28 1) 머신러닝의 활용···································································28 2) 머신러닝 알고D즘································································34 3) 예측 모델 학습 및 성능 평가 ·············································· 39 III. 연구 문제 및 방법···················································44 1. 연구 문제 ··············································································· 44 2. 연구 방법 ··············································································· 45 IV. BubbleUp 시스템 구현····················································47 1. 인터페이스 소개 ·································································· 48 1) 시스템 컨트롤 뷰 ······························································· 50 2) 기본 데이터 시각화 ····························································· 51 3) 클러스터링 시각화 ······························································· 52 4) 유사 데이터 시각화 ····························································· 53 5) 데이터의 전체 분포 및 상관관계 시각화 ···························· 55 6) 유사한 결과 분포 및 랭킹 리스트 인터페이스 ··················· 57 2. BubbleUp 시스템의 인터렉션··············································58 3. BubbleUp 시스템의 워크플로우 ········································ 59 1) 데이터 전처리······································································60 2) 머신러닝···············································································61 3) 데이터 분석··········································································62 4) 시각화···················································································63 4. 정리 ························································································· 64 V. 사례연구 ····································································· 67 1. 사례연구 1 – 치매 환자 검사 기록 데이터 ···················· 67 1) 시나리오···············································································68 2) 사례분석···············································································70 2. 사례연구 2 – NBA 데이터·················································76 1) 시나리오···············································································77 2) 사례분석···············································································79 VI. 시각화 사용성 평가 ··················································· 85 1. 연구설계 ················································································· 85 1) 자료 수집 및 표본 설정 ····················································85 2) 신뢰도 분석··········································································86 3) 문항 별 결과 비교 ······························································· 87 4) 사후 인터뷰 ·········································································· 90 2. 시각화에 대한 FGI·······························································91 1) BubbleUp 시각화의 목적 ·················································· 91 2) BubbleUp 시각화의 장점 ···················································· 92 3) BubbleUp 시각화의 단점 ···················································· 93 4) BubbleUp 시각화의 차별성 ················································ 94 5) BubbleUp 시각화의 추가 기능 및 개선점 ························· 95 VII. 결론··············································································97 1. 연구 요약 ··············································································· 97 2. 제언 ························································································· 99 참고문헌 ········································································ 100 부록 ················································································ 110 kor The Graduate School, Ajou University 아주대학교 논문은 저작권에 의해 보호받습니다. 버블업: 시간적 이벤트 데이터의 유사성 분석을 위한 시각화 시스템 GUO WENJUN Thesis 아주대학교 일반대학원 GUO WENJUN 일반대학원 라이프미디어협동과정 2018. 8 Master I804:41038-000000028017 http://dcoll.ajou.ac.kr:9080/dcollection/common/orgView/000000028017 데이터 시각화 유사성 분석 데이터 클러스터링 머신러닝 데이터 분석 Temporal event data is receiving increasing attention in a variety of fields. It is also helpful to identify a new configuration through comparison and analysis of patterns and similarities of events according to the change of time of temporal event data. Previous studies have focused on events that changed over time, but more and more research is being done on the changes in longitudinal time caused by repetition. In this research, an optimized "BubbleUp" visualization system for analyzing similarity of longitudinal temporal event data was proposed and developed. The visualization system was used to verify the following effects through case studies and usability evaluations. First, a visualization algorithm and user prototype of the BubbleUp system were implemented; furthermore, additional interaction functions and user interface were developed. User-friendly interaction has made it easier to identify similar relationships with other data to suit user needs. Second, the system facilitated the searching of data and it was possible to derive the potential pattern of data through clustering. Third, when the user selected the target, the system not only presented similar and dissimilar results but also freely set the similarity distribution range desired by the user. Similar results were also provided in the similarity result view with the similarity result rank list based on the set range. Fourth, it provided a means of comparing and analyzing future result effectively by predicting data through machine learning. Therefore, through this study, we confirmed that the BubbleUp system is a novel visualization system that intuitively enables integrated analysis of similarity, clustering, and prediction of longitudinal temporal event data.