# How to visualize large graphs with directed edge bundling

I'm wondering if anyone has implemented edge bundled graph plotting?

http://www.win.tue.nl/~dholten/papers/forcebundles_eurovis.pdf

-
While the paper is nice, can you expound upon it in your question? Links are fragile, so the more information that you add to your question, the longer they will remain relevant. –  rcollyer May 8 '12 at 0:44
Looks like a really intricate application of the EdgeRenderingFunction option to GraphPlot. –  Verbeia May 8 '12 at 1:27
@Verbeia it looks like more than that to me. It bundles edges that go in similar directions together, increasing their relative weight, and they peal off from the bundle near their nodes. (Fig. 8 is a good one to look at to see this.) This reduces the visual complexity of network. –  rcollyer May 8 '12 at 2:03

CommunityGraphPlot has this feature implemented internally:

g = ExampleData[{"NetworkGraph", "DolphinSocialNetwork"}];
CommunityGraphPlot[g]


The information about bundling is calculated by CommunityGraphPlot (i.e. it is not supplied with the example data), though unfortunately there are no documented options available to finetune the bundling (but see below), only the region (CommunityRegionStyle) and boundary styles (CommunityBoundaryStyle).

Though one can extract from the community graph the BezierCurve that generates the bundled edges (with low opacity EdgeStyle by default), it will contain references to vertex coordinates like DynamicLocation["VertexID$1", Automatic, Center]. Unfortunate again that due to the dynamical nature of a CommunityGraphPlot, ordinary Graph functions like VertexList or PropertyValue[{g, 1}, VertexCoordinates] will fail, so there is no easy way to extract the real vertex coordinates which otherwise could have been matched with the various "VertexID$1" references.

The method to find communities can be customized with FindGraphCommunities.

## Undocumented functionality

You have to dig deep to ultimately find any bundling-related code in GraphComputationGraphCommunitiesPlotDumpcommunitiesPlot. It accepts the following internal options (above default ones):

 "EdgeLayout" -> Automatic
"CommunityEdgeWeight" -> Automatic
"CommunityRegionFunction" -> Automatic


If "EdgeLayout" has the value "DividedEdgeBundling", the following suboptions are available to control the bundling process (with their defaults):

"CoulombConstant" -> 4.5
"VelocityDamping" -> 1
"SmoothEdge" -> True


Examining its internal usage and a bit of experimentation yielded the following results:

SetProperty[g, {GraphLayout -> {"EdgeLayout" -> {"DividedEdgeBundling",
"CoulombConstant" -> 5, "VelocityDamping" -> .6, "SmoothEdge" -> True},
"VertexLayout" -> {Automatic}}}]


SetProperty[g, {GraphLayout -> {"EdgeLayout" -> {"DividedEdgeBundling",
"CoulombConstant" -> 100, "VelocityDamping" -> .5, "SmoothEdge" -> False}}}]


SetProperty[g, {
GraphLayout -> {"EdgeLayout" -> {"DividedEdgeBundling",
"CoulombConstant" -> 44, "VelocityDamping" -> .87, "SmoothEdge" -> False}}}]


SetProperty[g, {
GraphLayout -> {"EdgeLayout" -> {"DividedEdgeBundling",
"CoulombConstant" -> 100, "VelocityDamping" -> .5, "SmoothEdge" -> False},
"VertexLayout" -> {"SpringEmbedding", "EdgeWeighted" -> True,
"Multilevel" -> False}}}]


For larger graphs it really helps to use transparent edges:

sm = SocialMediaData["Facebook", "FriendNetwork"];
{
SetProperty[sm, {ImageSize -> 300, GraphLayout -> {"EdgeLayout" -> None}}],
SetProperty[sm,
{ImageSize -> 300, EdgeStyle -> Opacity@.1,
GraphLayout -> {"EdgeLayout" -> {"DividedEdgeBundling",
"CoulombConstant" -> 500, "VelocityDamping" -> .5, "SmoothEdge" -> False}}}]
}


-