GION: Interactively Untangling Large Graphs on Wall-Sized Displays - Graph Drawing - page 117

the ability to stretch out parts of the graph layoutbyquickly moving different clusters to
new positions. A future extension to this techniquewould allow users to explicitly shrink
bones, bringing clusters together. This extension was not included in the evaluation.
We a l s o m a ke use of the clustering information to color the graph, based on the
degree of physics bodies. The clusters with the highest degree (most linked) are drawn
in red, and clusters with the lowest degree are colored green. The remaining clusters
are colored as a gradient from red to green based on their cohesion. Coloring the graph
in this way allows users to quickly identify which parts of the graph will be easiest to
move into more desirable locations.

4 Evaluation

We conducted a user study to evaluate the benefits of the physics-based graph inter-
action technique. We chose an untangling task for the experiment. Participants were
shown a series of graph layouts,andwereaskedto'untangle' the layout to better show
the overall structure. Untangling was chosen as the task because it would require many
mouse operations to complete, and resultinggraphs could be compared to the initial
layout. The experiment is a 2x2, within participant, repeated measures design. The con-
ditions tested were interaction mode: physics or normal ,andcoloring: colored or plain .
Participants used the GION techniqueforthe physics condition. The normal interaction
allowed users to move nodes to new locations one cluster at a time, emulating move-
ment operations commonly used in graph layout software.
The hypotheses tested in the experiment are as follows:

H1 Physics interaction leads to lower stress than normal interaction.
H2 Physics interaction leads to fewer edge crossingsthan normal interaction.
H3 Colored graphs would have lower stress than the plain graphs.
H4 Colored graphs would have fewer edge crossings than the plain graphs.
H5 Physics interaction requires less mouse movement than normal interaction.
H6 Physics interaction requires fewer clicks than normal interaction.
H7 Physics interaction leads to lower stress than the starting layout.
H8 Physics interaction leads to fewer edge crossings than the starting layout.
H9 Physics interaction is preferred by users.

The following data were collected during each trial of the experiment: mouse move-
ment, in millimetres on the videowall, mouse clicks, and snapshots of the graph layout
(captured every five seconds). From the graph layout snapshots other properties of the
layoutcould be calculated and analysed. Participants were asked to fill outaquestion-
naire at the end of the user study session. Participants answered questions 1-4 for both
the physics and normal interaction conditions, and questions 5-6 for the color condition.
All questions were answered using avisual analogue scale. The participants were asked
to rank the interaction conditions in order of preference, and comment on strategies
used for untangling the graphs.

1. Movinggraph clusters into new positions was { very easy - very hard }

2. Untangling the graphs was { very easy - very hard }