i
i
i
i
i
i
i
i
27.8. Examples 703
fied by using only straight edges, or by only allowing right-angle bends for the
class of orthogonal layouts, but some approaches handle true curves. If the graph
has directed edges, a layered approach can be used to show hierarchical struc-
ture through the horizontal or vertical spatial ordering of nodes, as shown in Fig-
ure 27.2.
Figure 27.22. Graph lay-
out aesthetic criteria. Top:
Edge crossings should be
minimized. Middle: Angular
resolution should be max-
imized. Bottom: Symme-
try is maximized on the left,
whereas crossings are min-
imized on the right, showing
the conflict between the in-
dividually NP-hard criteria.
A suite of aesthetic criteria operationalize human judgements about readable
graphs as metrics that can be computed on a proposed layout (Ware et al., 2002).
Figure 27.22 shows some examples. Some metrics should be minimized, such
as the number of edge crossings, the total area of the layout, and the number of
right-angle bends or curves. Others should be maximized, such as the angular
resolution or symmetry. The problem is difficult because most of these criteria
are individually NP-hard, and moreover they are mutually incompatible (Bran-
denburg, 1988).
Many approaches to node-link graph drawing use force-directed placement,
motivated by the intuitive physical metaphor of spring forces at the edges drawing
together repelling particles at the nodes. Although naive approaches have high
time complexity and are prone to being caught in local minima, much work has
gone into developing more sophisticated algorithms such as GEM (Frick et al.,
1994) or IPSep-CoLa (Dwyer et al., 2006). Figure 27.23 shows an interactive
system using the r-PolyLog energy model, where a focus+context view of the
clustered graph is created with both geometric and semantic fisheye (van Ham &
van Wijk, 2004).
Figure 27.23. Force-directed placement showing a clustered graph with both geometric
and semantic fisheye.
Image courtesy Jarke van Wijk
(van Ham & van Wijk, 2004),
c
2004
IEEE.