Title Page
Abstract
Contents
Chapter 1. Introduction 20
1.1. Background and Motivation 20
1.2. Research Questions 24
1.3. Contributions 24
1.3.1. Augmenting Parallel Coordinates Plots with Color-coded Stacked Histograms 25
1.3.2. Interactive Visualization System for Monitoring Issues in Industrial Sofware Development 26
1.3.3. Towards Supporting the Exploration of Temporal Rank Data with Multiple Colormaps 27
1.4. Structure of the Dissertation 28
Chapter 2. Related Work 30
2.1. Colors as a Visual Channel 30
2.2. Approaches in Visual Information Seeking 32
2.3. Approaches in Visualizing with Scalability 34
2.3.1. Parallel Coordinate Plots 34
2.3.2. Visualizing Temporal Rank Data 37
Chapter 3. Augmenting Parallel Coordinates Plots with Color-coded Stacked Histograms 40
3.1. Background 40
3.2. Design of PHP 42
3.2.1. Design Rationale 42
3.2.2. Construction of Color-coded Histograms 44
3.3. Visual Information Seeking with PHP 46
3.3.1. Interpreting PHP 47
3.3.2. Tools for Zooming in on Small bars 51
3.4. Use Case 54
3.5. User Study 59
3.5.1. Design 60
3.5.2. Results 63
3.6. Discussion 69
3.7. Summary 72
Chapter 4. Interactive Visualization System for Monitoring Issues in Industrial Soware Development 73
4.1. Motivation 73
4.2. Background 74
4.2.1. Interview 74
4.2.2. Domain Situation 75
4.2.3. Task 77
4.3. Issue Data 79
4.4. IssueML 81
4.4.1. Attribute View 83
4.4.2. Issue List View 84
4.4.3. Analysis View 84
4.4.4. Information View 87
4.4.5. Issue Analysis View 88
4.5. Use Case 91
4.6. Discussion 93
4.7. Summary 96
Chapter 5. Towards Supporting the Exploration of Temporal Rank Data with Multiple Colormaps 97
5.1. Background 97
5.2. Design Rationale 100
5.3. TRaVis 101
5.3.1. Rendering of TRaVis 101
5.3.2. Controlling Visual Components 104
5.4. Use Case 108
5.4.1. Music Rank Data 109
5.4.2. Ranks of Popular Baby Names 112
5.4.3. National Soccer Team Rankings 113
5.5. Discussion 116
5.6. Summary 120
Chapter 6. Discussion 121
6.1. Lessons Learned 121
6.1.1. Significance of Utilizing Colors in Expressing Multiple Values 121
6.1.2. Bridging the Gaps in Information Visualization 125
6.2. Limitations 129
Chapter 7. Conclusion 131
7.1. Summary of Contributions 131
7.2. Future Research Agendas 133
7.2.1. Expanding the Interaction with Multiple Values using Colors 133
7.2.2. Assessing the Effectiveness and Scalability in Color Patterns 134
7.2.3. Further Supporting Novices with Visualizations 135
7.3. Final Remarks 136
Bibliography 138
국문 초록 146
Table 4.1. Major fields in the issues utilized for IssueML. 79
Table 4.2. Structure of each record consisting the overall log data 80
Table 4.3. Classification of issues according to a developer 81
Figure 1.1. Parallel Histogram Plots (PHP) visualizing multiple attributes. 25
Figure 1.2. Overview of IssueML displaying multiple issues. 26
Figure 1.3. TRaVis visualizing the ranking changes of multiple items. 27
Figure 3.1. Parallel Histogram Plots (PHP) used to draw the CASP dataset [78]. The attribute F2 is selected for color coding. From the color distribution, it can be deduced that... 42
Figure 3.2. How PHPis constructed from data. (a)Data items are sorted by a user-selected attribute (attribute B), and items are grouped (Ga-Gd) according to the sorted order, i.e.,...[이미지참조] 46
Figure 3.3. The Baseball dataset [18] rendered in PHP. The attribute selected for color coding (i.e., the pivot attribute) is bordered by a green rectangle ((a) IP and (b) wFB). Re-... 47
Figure 3.4. Example of positive and negative correlations displayed in PHP. The attributes, Positive and Negative have positive (+0.8) and negative (-0.8) correlations with... 48
Figure 3.5. Histogram of WAR from Figure 3.3(b). The area inside the green box is magnified for visibility. In the region of the green box, it can be observed that some blue items... 49
Figure 3.6. Displaying selected items in PHP. (a) Data items with high IP values are selected from Figure 3.3. The distribution of selected items can be compared with the over-... 49
Figure 3.7. Various interactions of PHP designed to deal with small components. (a) Focus+context zooming enables users to enlarge histograms of interest (the histograms of... 51
Figure 3.8. CASP dataset rendered in (a) PCP, (b) AH, and (c) PHP. F7 is set as the pivo attribute in PHP. In PHP, the relationship between F7 and other attributes can be discov-... 55
Figure 3.9. CASP dataset rendered as a scatter plot matrix (SPLOM) with the colors of PHP (F7 is the pivot attribute) applied to the scatter plots in the lower-right triangle of the ma-... 57
Figure 3.10. Results regarding task completion time in the first task. (a) Results by visualization method and correlation coefficient set. Error bars indicate the standard deviation... 64
Figure 3.11. Results regarding error rate in the first task. (a) Result by visualization method and correlation coefficient set. Error bars indicate the standard deviation of the measured... 65
Figure 3.12. Performance evaluation results of the second task. Error bars indicate the standard deviation of the measured mean. (a) Response time of the visualization by posi-... 67
Figure 4.1. Diagram depicting the life cycle of issues. Issues are generated according to planned product tests, and allocated to developers for resolution. Managers monitor subset of the issues related to them, and provide feedbacks to future tests and aid the... 76
Figure 4.2. IssueML visualizing a real world data. Some fields are censored due to privacy. (a) From the attribute view, users can observe the distribution of each attribute, and select ranges or values of interest to filter the issues. (b) Each of the selected issues is... 82
Figure 4.3. Example of the filtering interaction applied in the attribute view. Upon selecting a certain range/categories in the attributes, the distribution of the selected issues can... 83
Figure 4.4. Step-to-step diagram of visualizing a single issue's information over time in the issue log component. (a) The overall visualization is a timeline, in which colors indi-... 86
Figure 4.5. Example of the threading approach in the comments fro man issue in the information view. In comment 3, Developer C mentions Developer Dand E, and in comment... 87
Figure 4.6. Overview of the issue analysis view. Actual values in the fields are screened out for privacy reasons. In issue analysis view, users can observe the detailed information of the single issue, and interact with similar issues (upper right). Users can also process the... 90
Figure 4.7. When a candidate from the list of similar issue is selected, the selected issue in displayed side to side, and the tokens that are shared in the issues are highlighted for... 91
Figure 4.8. Example of interacting with multiple issues in IssueML. In the Figure, a total of 1000 issues generated during the development of smart TVs are displayed. Users can... 93
Figure 4.9. Another example of interacting with multiple issues in IssueML in the issue log component. Among the patterns of multiple issues, users can discover distinct patterns... 94
Figure 5.1. The music rank chart [melon] visualized as TRaVis. In TRaVis, ranking changes of each items are displayed as rows of colormaps, enabling the visualization of multiple items in limited space without cluttering. The stacked color patterns reveal interesting... 98
Figure 5.2. How TRaVis is rendered from temporal rank data. (a) Ranks of 2 items A and B. Dash (-) indicates no records at that time. (b) Data of (a) visualized as line charts. In line... 102
Figure 5.3. Various interactions in TRaVis for supporting the observation of multiple items. (a) The border of each items can be toggled (top: with borders, bottom: without... 105
Figure 5.4. The music rank chart [melon] visualized as TRaVis,with border removed and blanks filled in with other items. Even though the information of each of the items is harder to be observed, the overall pattern of multiple items can be better displayed. 107
Figure 5.5. The music rank data visualized as line chart. Outside of the information of density generated from the cluttering of lines, it is hard for users to further interact with... 108
Figure 5.6. The upper right portion of TRaVis from Figure 5.1. Users can readily discover irregular patterns (black box) or items with repeated patterns (orange box) from the color... 109
Figure 5.7. Different sorting and filtering applied in the music rank data. Interactions in TRaVis can help discovery information in multiple perspectives. (a) Items ordered by... 111
Figure 5.8. Items filtered by the same artist of the black boxed song from Figure 5.7. It can be observed that the rise in rankings correspond to periodic appearances of new songs. 111
Figure 5.9. The Baby Name Data [4] rendered as TRaVis. Color distribution reveals different characteristics of the two dataset. (a-b) Names sorted by frequency of appearance... 112
Figure 5.10. The soccer ranking data [19] rendered as TRaVis. Changing the order of items in TRaVis is a critical interaction which allows the observation of items in novel perspec-... 114
Figure 5.11. By combining and altering the color scheme, the temporal rank data can be visualized in different perspectives in additional to the ranking values. Items in the music rank data [53] are ordered and colored by the genre of the songs, and the opacity is... 119