Network-based. Visual Analysis of Tabular Data. Zhicheng Liu, Shamkant Navathe, John Stasko

Transcription

1 Network-based Visual Analysis of Tabular Data Zhicheng Liu, Shamkant Navathe, John Stasko

2 Tabular Data 2

3 Tabular Data Rows and columns Rows are data cases; columns are attributes/dimensions Attribute types o Quantitative (numbers) o Ordinal (e.g. small, medium, large) o Nominal (names, categories) 3

4 Insight Discovery on Tabular Data: Example NSF Grants o Grant Title o Amount o Date o Program Manager o Awardee / Researcher Name o Awardee Affiliation o 4

5 Visualizing Tabular Data [Rao and Card, 1994] [Tableau] [Spotfire] 5

6 # Grants in each program 6

7 Amount by date 7

8 Amount by date & ProgMgr 8

9 Collaboration between institutions? 9

10 Relationship between ProgMgr and Researchers? 10

11 Quantitative attributes o Nominal data as independent variables, or not handled Patterns of distributions, correlations and outliers of numerical values Nominal attributes o Quantiative attributes are useful too Entities with interesting roles, emergent global structures 11

12 Current State of the Art Tabular Data Explicit Network Spotfire, Tableau, TableLens. GUESS, UCINet, SocialAction. 12

13 Problems with this Partition Analytical: Network semantics are dynamic Usability: Modeling networks is tedious and requires programming skill Counter-intuitive for Exploratory Analysis 13

14 NodeXL [Hansen, et. al. 2010] 14

15 Questions & Approaches Goal: Network-based Visual Analysis of Tabular Data 1. Which conceptually meaningful operations are necessary to extract and transform tabular data into networks for exploratory analysis? Domain independent Generalized operations Expressive power 2. Given a set of operations, how to provide analysts with easy access to these operations and to couple network modeling with exploratory analysis? Hide technical details Reduce articulatory distance Immediate visual feedback 15

16 Formal Framework Tables: Relational model [Codd, 1969] o Each row is uniquely identifiable o Values in each cell is atomic: number, boolean, string, date Networks: Weighted Simple Graphs o Undirected A o At most one edge between any two nodes o Edges are weighted B 16

17 An Example ID LastNm FirstNm Type Date Size Visitee Loc 1 Dodd Chris VA 6/25/ POTUS 2 Smith John VA 6/26/ Smith John AL 6/26/ Hirani Amyn VA 6/30/ Keehan Carol VA 6/30/ Keehan Carol VA 7/8/09 26 Office Visitors Amanda Kepko Office Visitors Kristin Sheehy Daniella Leger 17

18 First-order Graph: Single Table ID LastNm FirstNm Type Date Size Visitee Loc 1 Dodd Chris VA 6/25/ POTUS 2 Smith John VA 6/26/ Office Visitors 3 Smith John AL 6/26/ Amanda Kepko 4 Hirani Amyn VA 6/30/ Office Visitors 5 Keehan Carol VA 6/30/09 8 Kristin Sheehy 6 Keehan Carol VA 7/8/09 26 Daniella Leger 18

19 First-order Graph: Single Table ID LastNm FirstNm Type Date Size Visitee Loc 1 Dodd Chris VA 6/25/ POTUS 2 Smith John VA 6/26/ Office Visitors 3 Smith John AL 6/26/ Amanda Kepko 4 Hirani Amyn VA 6/30/ Office Visitors 5 Keehan Carol VA 6/30/09 8 Kristin Sheehy 6 Keehan Carol VA 7/8/09 26 Daniella Leger LastNm, Dodd Smith Smith Hirani Keehan Keehan FirstNm Chris John John Amyn Carol Carol Loc 19

20 First-order Graph: Single Table ID LastNm FirstNm Type Date Size Visitee Loc 1 Dodd Chris VA 6/25/ POTUS 2 Smith John VA 6/26/ Office Visitors 3 Smith John AL 6/26/ Amanda Kepko 4 Hirani Amyn VA 6/30/ Office Visitors 5 Keehan Carol VA 6/30/09 8 Kristin Sheehy 6 Keehan Carol VA 7/8/09 26 Daniella Leger LastNm, Dodd Smith Smith Hirani Keehan Keehan FirstNm Chris John John Amyn Carol Carol Loc 20

21 First-order Graph: Single Table ID LastNm FirstNm Type Date Size Visitee Loc 1 Dodd Chris VA 6/25/ POTUS 2 Smith John VA 6/26/ Office Visitors 3 Smith John AL 6/26/ Amanda Kepko 4 Hirani Amyn VA 6/30/ Office Visitors 5 Keehan Carol VA 6/30/09 8 Kristin Sheehy 6 Keehan Carol VA 7/8/09 26 Daniella Leger [Type] LastNm, FirstNm Loc [Type = VA] Dodd Chris [Type = VA] Smith John [Type = AL] Smith John [Type = VA] Hirani Amyn [Type = VA] Keehan Carol [Type = VA] Keehan Carol 21

22 Higher-order Graph: Transformations Aggregation Projection Edge Weighting Slicing n Dicing 22

23 Aggregation: Entity Resolution original graph after aggregation Dodd, Chris Smith, John Smith, John Hirani, Amyn Keehan, Carol Keehan, Carol Dodd, Chris Smith, John Smith, John Hirani, Amyn Keehan, Carol 23

24 Aggregation: Pivoting original graph after aggregation Dodd, Chris Smith, John VA Smith, John Hirani, Amyn Keehan, Carol AL Type Location 24

25 Projection Dodd, Chris Smith, John Hirani, Amyn Dodd, Chris Smith, John Keehan, Carol Hirani, Amyn Keehan, Carol 25

26 Edge Weighting ID LastNm FirstNm Type Date Size Visitee Loc 1 Dodd Chris VA 6/25/ POTUS 2 Smith John VA 6/26/ Office Visitors 3 Smith John AL 6/26/ Amanda Kepko 4 Hirani Amyn VA 6/30/ Office Visitors 5 Keehan Carol VA 6/30/09 8 Kristin Sheehy 6 Keehan Carol VA 7/8/09 26 Daniella Leger LastNm, Dodd Smith Hirani Keehan FirstNm Chris John Amyn Carol Visitee POTUS Office Visitors Amanda Kepko Kristin Sheehy Daniella Leger 26

27 Edge Weighting ID LastNm FirstNm Type Date Size Visitee Loc 1 Dodd Chris VA 6/25/ POTUS 2 Smith John VA 6/26/ Office Visitors 3 Smith John AL 6/26/ Amanda Kepko 4 Hirani Amyn VA 6/30/ Office Visitors 5 Keehan Carol VA 6/30/09 8 Kristin Sheehy 6 Keehan Carol VA 7/8/09 26 Daniella Leger LastNm, FirstNm Visitee Dodd Chris 2018 POTUS Smith John 237 Office Visitors Hirani Amyn Amanda Kepko Keehan Carol 8 Kristin Sheehy 26 Daniella Leger 27

28 Slice n Dice ID LastNm FirstNm Type Date Size Visitee Loc 1 Dodd Chris VA 6/25/ POTUS 2 Smith John VA 6/26/ Office Visitors 3 Smith John AL 6/26/ Amanda Kepko 4 Hirani Amyn VA 6/30/ Office Visitors 5 Keehan Carol VA 6/30/09 8 Kristin Sheehy 6 Keehan Carol VA 7/8/09 26 Daniella Leger LastNm, Dodd Smith Hirani Keehan FirstNm Chris John Amyn Carol Visitee POTUS Office Visitors Amanda Kepko Kristin Sheehy Daniella Leger 28

29 Slice n Dice ID LastNm FirstNm Type Date Size Visitee Loc 1 Dodd Chris VA 6/25/ POTUS 2 Smith John VA 6/26/ Office Visitors 3 Smith John AL 6/26/ Amanda Kepko 4 Hirani Amyn VA 6/30/ Office Visitors 5 Keehan Carol VA 6/30/09 8 Kristin Sheehy 6 Keehan Carol VA 7/8/09 26 Daniella Leger LastNm, FirstNm Visitee LastNm, FirstNm Visitee Dodd Chris POTUS Smith John Office Visitors Smith John Amanda Kepko Hirani Amyn Keehan Carol Kristin Sheehy Keehan Carol Daniella Leger 29

30 Expressive Power Proximity grouping Extending to directed one-mode network Limitations 30

31 Ploceus Interface Overview Data Management View Network Visualization View Network Schema View 31

32 Direct Manipulation Interface 32

33 Ploceus Demo 33

34 Related Work Centrifuge Orion 34

35 Multiple Tables GID Title Program Data Mining of Digital Behavior Real-time Capture, Management and Reconstruction of Spatio-Temporal Events Statistical Data Mining of Time-Dependent Data with Applications in Geoscience and Biology Statistics Information Technology Research ITR for National Priorities Program Manager Sylvia Spengler Maria Zemankova Sylvia Spengler Amount Year PID Name Org 1 2 Padhraic Smyth Sharad Mehrotra Person Grant Role 1 1 PI 2 1 copi 2 2 PI University of California Irvine University of California Irvine 1 3 PI 35

36 First-order Graph: Multiple Tables GID Title Program ProMgr Amount Year 1 Data Mining of Digital Sylvia Statistics Behavior Spengler Real-time Capture, Information Tech- Maria Management nology Research Zemankova Statistical Data Mining ITR for National Sylvia of Time-Dependent Priorities Spengler PID Name Org 1 Padhraic Smyth University of California Irvine 2 Sharad Mehrotra University of California Irvine Grant Role Person 1 PI 1 1 copi 2 2 PI 2 3 PI 1 Title Program ProMgr Amount Year Grant Role Person PID Name Org 1 PI 1 1 copi 2 2 PI 2 3 PI 1 36

37 Open Issues with Multiple Tables (1) Join Specification 37

38 Open Issues with Multiple Tables (2) Interpretation of Edge Weights 38

39 Acknowledgments IIS VACCINE Center 39