Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Relaxed Selection Techniques for Querying Time-Series Graphs

Relaxed Selection Techniques for Querying Time-Series Graphs

Time-series graphs are often used to visualize phenomena that change over time. Common tasks include comparing values at different points in time and searching for specified patterns, either exact or approximate. However, tools that support time-series graphs typically separate query specification from the actual search process, allowing users to adapt the level of similarity only after specifying the pattern. We introduce relaxed selection techniques, in which users implicitly define a level of similarity that can vary across the search pattern, while creating a search query with a single-gesture interaction. Users sketch over part of the graph, establishing the level of similarity through either spatial deviations from the graph, or the speed at which they sketch (temporal deviations). In a user study, participants were significantly faster when using our temporally relaxed selection technique than when using traditional techniques. In addition, they achieved significantly higher precision and recall with our spatially relaxed selection technique compared to traditional techniques.

More information on http://www.christianholz.net/relaxed_selection_techniques.html

Christian Holz

October 06, 2009
Tweet

More Decks by Christian Holz

Other Decks in Research

Transcript

  1. .&/0/()1/$(/#) <*:"+(1*)(%$=%>$1:.(*+%?5-*)5*% >$'.1@-"%A)-B*+#-(42%C*D%E$+F2%CE% =*-)*+95#75$'.1@-"7*0.% 3#$-4%/1(!*-$%./!(%/(#($%3*5!*)%*!(1)#"4 ABSTRACT Time-series graphs are often

    used to visualize phenomena that change over time. Common tasks include comparing values at different points in time and searching for speci- fied patterns, either exact or approximate. However, tools that support time-series graphs typically separate query specification from the actual search process, allowing users !"#$%#&&'()*&#+*,(!*&*-$%./ )*!0&$%/1(!*&*-$%./(2($.&*)#/-*! $*3".)#&&'()*&#+*,(!*&*-$%./ !6*$-4*,("#$$*)/! Figure 1: Relaxed selection techniques. Spatially relaxed s between displayed graph and user sketch on a point-by-po are derived from input speed. (Center) In both cases, toler lection. (Right) Tolerant selections allow similarity matches ABSTRACT Time-series graphs are often used to visualize phenomena that change over time. Common tasks include comparing values at different points in time and searching for speci- fied patterns, either exact or approximate. However, tools that support time-series graphs typically separate query specification from the actual search process, allowing users INTRODUCTION Time-series graphs are types of diagram for v time to a single spatial ax naïve viewer to compre time and visually search portions of the graph. W !"#$%#&&'()*&#+*,(!*&*-$%./ )*!0&$%/1(!*&*-$%./(2($.&*)#/-*! $*3".)#&&'()*&#+*,(!*&*-$%./ 3#$-4%/1(!*-$%./!(%/(#($%3*5!*)%*! !6*$-4*,("#$$*)/! Figure 1: Relaxed selection techniques. Spatially relaxed selection (top): Tolerances are between displayed graph and user sketch on a point-by-point basis. Temporally relaxed are derived from input speed. (Center) In both cases, tolerances can be visualized with lection. (Right) Tolerant selections allow similarity matches for patterns that recur in a tim 1 Relaxed Selection Techniques for Querying Time-Series Graphs Christian Holz Steven Feiner mbia University 2008–2009 es and Payments Columbia University 2008–2009 Guide to Fees and Payments Columbia University 2008–2009 Guide to Fees and Payments query matches selection 2 3
  2. 0 15 30 45 60 2006/II Untitled 25 Untitled 32

    Untitled 39 Untitled 46 Untitled 53 Untitled 60 Untitled 67 2008/II Untitled 81 Untitled 88 Untitled 95 Untitled 102 0 15 30 45 60 2006/II 2007/I 2007/II 2008/I 2008/II 2009/I rubber-band selection
  3. 0 15 30 45 60 2006/II Untitled 25 Untitled 32

    Untitled 39 Untitled 46 Untitled 53 Untitled 60 Untitled 67 2008/II Untitled 81 Untitled 88 Untitled 95 Untitled 102 0 15 30 45 60 2006/II 2007/I 2007/II 2008/I 2008/II 2009/I 0 15 30 45 60 2006/II Untitled 25 Untitled 32 Untitled 39 Untitled 46 Untitled 53 Untitled 60 Untitled 67 2008/II Untitled 81 Untitled 88 Untitled 95 Untitled 102 rubber-band selection similarity
  4. 0 15 30 45 60 2006/II Untitled 25 Untitled 32

    Untitled 39 Untitled 46 Untitled 53 Untitled 60 Untitled 67 2008/II Untitled 81 Untitled 88 Untitled 95 Untitled 102 0 15 30 45 60 2006/II 2007/I 2007/II 2008/I 2008/II 2009/I rubber-band selection
  5. 0 15 30 45 60 2006/II Untitled 25 Untitled 32

    Untitled 39 Untitled 46 Untitled 53 Untitled 60 Untitled 67 2008/II Untitled 81 Untitled 88 Untitled 95 Untitled 102 0 15 30 45 60 2006/II 2007/I 2007/II 2008/I 2008/II 2009/I 0 15 30 45 60 2006/II Untitled 25 Untitled 32 Untitled 39 Untitled 46 Untitled 53 Untitled 60 Untitled 67 2008/II Untitled 81 Untitled 88 Untitled 95 Untitled 102 rubber-band selection similarity
  6. TimeSearcher Hochheiser and Shneiderman, Discovery Science’01. Keogh et al., FQAS'02.

    Hochheiser et al. ICME'03. Hochheiser and Shneiderman, Information Visualization’04 Buono et al., VDA’05. Buono and Simeone, AVI’08.
  7. Relevance Feedback Keogh and Pazzani, SIGIR’99 Patterns Morrill, Communic. ACM’98

    VizTree Lin et al., KDD'04 Lin et al., VLDB'04 Ryall et al., CHI’05 QueryLines Line Graph Explorer Kincaid and Lam, AVI’06
  8. design each participant: 4 techniques × 9 datasets × 15

    interactions _________________________ = 540 trials
  9. results: exact matches !"!!# !"$!# !"%!# !"&!# !"'!# !"(!# !")!#

    !"*!# +,-.#,/-01# precision recall 0 % 10 % 20 % 30 % 40 % 50 % 60 % 70 % SRS TRS RB QE SRS TRS RB QE
  10. results: exact matches !"!!# !"$!# !"%!# !"&!# !"'!# !"(!# !")!#

    !"*!# +,-.#,/-01# precision recall 0 % 10 % 20 % 30 % 40 % 50 % 60 % 70 % SRS TRS RB QE SRS TRS RB QE
  11. results: intersecting matches +,-.#123,42# precision recall 0 % 10 %

    20 % 30 % 40 % 50 % 60 % 70 % SRS TRS RB QE SRS TRS RB QE
  12. results: intersecting matches +,-.#123,42# precision recall 0 % 10 %

    20 % 30 % 40 % 50 % 60 % 70 % SRS TRS RB QE SRS TRS RB QE
  13. results: exact matches +,-.#,/-01#5341# precision recall 0 % 10 %

    20 % 30 % 40 % 50 % 60 % 70 % SRS TRS RB QE SRS TRS RB QE rst interaction
  14. results: exact matches +,-.#,/-01#5341# precision recall 0 % 10 %

    20 % 30 % 40 % 50 % 60 % 70 % SRS TRS RB QE SRS TRS RB QE rst interaction
  15. results: intersecting matches +,-.#123,42#5341# precision recall 0 % 10 %

    20 % 30 % 40 % 50 % 60 % 70 % SRS TRS RB QE SRS TRS RB QE rst interaction
  16. results: intersecting matches +,-.#123,42#5341# precision recall 0 % 10 %

    20 % 30 % 40 % 50 % 60 % 70 % SRS TRS RB QE SRS TRS RB QE rst interaction
  17. !" #!!" $!!!" $#!!" %!!!" %#!!" &'(')*+," -./0&12',1" 0 s

    0.5 s 1.0 s 1.5 s 2.0 s results: time selection adjustment SRS TRS RB QE SRS TRS RB QE
  18. !" #!!" $!!!" $#!!" %!!!" %#!!" &'(')*+," -./0&12',1" 0 s

    0.5 s 1.0 s 1.5 s 2.0 s results: time selection adjustment SRS TRS RB QE SRS TRS RB QE
  19. !" #!!" $!!!" $#!!" %!!!" %#!!" &'(')*+," -./0&12',1" 0 s

    0.5 s 1.0 s 1.5 s 2.0 s results: time selection adjustment SRS TRS RB QE SRS TRS RB QE
  20. conclusions 2 new techniques SRS: spatially relaxed selection TRS: temporally

    relaxed selection study: TRS fastest for selection. SRS found the most relevant matches.
  21. Q?