Folded Timelines
Note
From now on, this page will not be updated anymore. A Google document is created for further updating: google doc link
Participants
- Shi Conglei (main investigator)
- Pierre Dragicevic
- Petra Isenberg
- Jean-Daniel Fekete
Summary
This project investigates techniques for bending/curving linear visualizations (e.g., timelines) to encode extra information.
There are many ways of doing this but a particularly interesting technique is to first lay out a sequence of points (ordered events) on a linear timeline so that the distance between two successive points ei, ei+1 is proportional not to their time difference, but some other quantity X(ei, ei+1). After doing this we curve the timeline so that the distance between any two points ei, ej is proportional to X(ei, ej). This is very similar to MDS (multidimensional scaling), except MDS techniques not assume points connected with a curve.
One application of this is giving rapid overviews of the "shape" of Wikipedia articles edit histories. X would then be the edit distance. If the overall shape is wiggly, then there has been many disagreements and the article has been edited a lot but did not progress very fast in practice because of many backtracks. If the shape of the curve is close to linear, then the editing process has been mostly additive (like an article listing very factual things). See Pierre's Java prototype for concrete examples.
This overview technique can be applied to any constructive process, like coding or writing an article, either alone or in groups. We just need a distance metrics. We need a proper distance metrics because if, e.g., the triangular inequality property is not met then there would be cases where we'd have to tear the timeline apart.
Key challenges in this project are to:
- Find other datasets to demonstrate the usefulness of the technique
- Position this technique with respect to related work
- Make the timeline interactive, e.g., curvilinear pan-and-zoom navigation, details-on-demand, dynamic update, etc.
Other Resources
- Pierre's Java prototype
- SVN URL: https://scm.gforge.inria.fr/svn/aviz/foldedcurves/
- Main class: fr.aviz.foldedcurves.sandbox.physics.Main
- Data set loading: fr.aviz.foldedcurves.sandbox.physics.WikipediaMatrix
- Other docs (possibly move the content here)
- notes.txt in the svn
- todo.txt in the svn
Related Work
- Multidimensional scaling
- (JDF) Sheppard Diagrams
- (Yvonne) Fly
- (Petra) http://www.cabinetmagazine.org/issues/13/timelines.php. Scroll to 1767
- (Benjamin) Liao and al., Visualizing Graph Dynamics and Similarity for Enterprise Network Security and Management. See image page 4.
- (Benjamin) Action synopsis
- (Benjamin) http://mbostock.github.com/protovis/ex/driving.html
- Karan Singh's space-filling curves
Possible Datasets
- Wikipedia (partly implemented)
- Articles and code from Aviz SVN
- See datasets used in Arc Diagrams
Ideas (with acknowledgements)
- (JDF) local compression of timeline, multidimensional scaling, scaling up
- (Stéphane Huot) Context (blocked article, news, Bush, etc.)
- (Anastasia) local compression of timeline
- (Mountaz) local aggregation (semantic: edit wars, etc.), 2D footprint, scaling up
- (Petra) timeline vs. non-linear timeline, showing time (pitching meeting)
- (Evelyne) evolutionary algorithms to optimize folding