GraphDice - Examples of use

Filter by value

A constant value is one of the automatically computed dimension. It is helpful to better see the repartition of the actors according to one specific dimension, and the connections between the per-value groups (the use of the excentric label or selections helps emphasize the connection patterns). It makes it also easier to perform a lasso selection for a per-value selection.
This example shows the Infovis conference co-authorship network. It contains the first date and last date of publication of each researcher in the network, at the InfoVis conference. One can compare patterns of interest:

  • Switching to the "first date" x "CST" plot, and activating the jitter makes it possible to see the most popular year as a first InfoVis publication year (1995).
  • Filtering out by value (1995) is performed using lasso selection.
  • To compare with the latest publication pattern, switch to the "last date" x "CST" view. The most popular year as a last InfoVis publication is easy to identify and select (1998), using a different query color.

Directed links

In GraphDice, directed links are drawn using biased splines, where bias and orientation express the link direction.

This example shows migration paths (edges) between cities (nodes) in northern France. Here's how one can compare the reciprocity patterns of the most central city (y-axis) for each of the cities' dominant language (x-axis: respectively Flemish, both, and French):

  • Select the first city of interest.
  • Make sure the selection's outlinks to be activated so that the node-related pattern is emphasized as the outgoing links are colored.
  • The most central Flemish-speaking city (WATGOERSERV) reveals to have a reciprocal migrations pattern, mainly between cities of the same dominant language, as shows the petal shape of related links on the same x-value.
  • Clear the selection so that it is possible to focus on the second city by selecting it in turn.
  • The second city (VXBERQUIN), that has both French and Flemish as a dominant language has bi-directionnal migrations with various cities, no matters their dominant language.
  • The third city (LILLE) has a different pattern as it only welcomes immigrant.
  • The history of selection makes it easy to quickly get back to any patterns by just hovering over the 3 latest states.


Axes can encode intervals, i.e. value ranges. Intervals allow aggregatation of numerical attributes of actors, e.g., the years of a researcher's publications can be replaced with the year of her first and last publications. GraphDice visualizes intervals as straight lines with short hooks in the ends, to distinguish them from links between actors.
This example shows the co-authorship network of INRIA project-teams for four years. By looking at the network, one can easily see that lots of projects don't have any connections (single nodes on the external circle). One can select those teams and perform further exploration. The first hypothesis is that the non-connected teams are too recent for having published yet.

  • The teams of interest correspond to the red selection, make sure the queries are visible using the 3-states buttons - Switch to the interval view showing the years of publication
  • The teams of interest are spread out in the new plot:
    • (1) some of them are one year old at most (2008 dataset) so they are still too young projects to worry about,
    • (2) other ones are already finished projects, so that it's no-sense to keep worrying about and finally,
    • (3) two team-projects remains.
  • By using the excentric label lens, the head of the research institute can identify the two teams for further analysis to figure out why they've not collaborated yet.