Information Technology Reference
In-Depth Information
Ta b l e 1 .
The mean values and the pairwise significance between each pair of drawing paradigms
are shown for error rate and for response time considering all the tasks (Overall) and single tasks
Error rate
Response time (s)
Overall
Overall
PA
PA
DE
CA
CY
DE
CA
CY
mean OOD
0.119
0.190
0.047 0.154
0.083
58
78
36
69
48
mean HD
0.199
0.250
0.142 0.190
0.214
55
69
41
55
56
mean OD
0.369
0.547
0.095 0.369
0.452
67
114
17
71
66
mean MR
0.423
0.559
0.047 0.285
0.809
129
187
27
145
158
OOD vs HD
.003
n.s.
n.s.
n.s.
0.008
n.s.
n.s.
n.s.
.002
n.s.
OOD vs OD
<.
001
.001
n.s.
.003
<.
001
n.s.
n.s.
<.
001
n.s.
n.s.
OOD vs MR
<.
001
<.
001 n.s
n.s.
<.
001
<.
001 .001
n.s.
<.
001
<.
001
HD vs OD
<.
001
.002
n.s.
.003
.001
n.s.
.003
<.
001
.006
n.s.
HD vs MR
<.
001
.001
n.s.
n.s.
<.
001
<.
001 .001
.001
<.
001
<.
001
OD vs MR
n.s.
n.s.
n.s.
n.s.
<.
001
<.
001 .002
.001
<.
001
<.
001
Experiments.
We chose 4 different graphs, modeling both real and artificial networks,
with and without cycles, with size (number of vertices) varying in the range [77
,
122]
and density in the range [2
.
5
,
3
.
5]; for each graph we computed 4 drawings using the
yEd Graph Editor
1
implementations of HD and OD and our own implementations of
OOD and MR. After a training session, for each drawing the participants had to solve 4
tasks:
(PA)
“Is there a path between the two highlighted vertices?”;
(DE)
“What is the
out-degree of the highlighted vertex?”;
(CA)
“Dothetwohighlighted vertices have any
common adjacent vertex?”;
(CY)
“Is there a cycle including the highlighted vertex?”.
We compared the performance of all the drawing paradigms in terms of error rate and
response time. 21 volunteering students participated in the experiments. We performed
a non parametric analysis, the results are summarized in Table 1.
Conclusions.
The results show a clear advantage in terms of accuracy in the reading
of the displayed graphs when using the OOD paradigm, over all tasks and in particular
for the tasks involving paths (PA) and cycles (CY). In terms of response time, the per-
formance on the node-link representations (OOD, HD, OD) are comparable, although
most tasks are executed slightly faster using HD. MR led to slower performance, ex-
cept for task DE. In addition, node-link representations outperformed the matrix-based
representation, both in terms of error rate and response time, especially for task CY.
References
1. Ghoniem, M., Fekete, J.-D., Castagliola, P.: A comparison of the readability of graphs using
node-link and matrix-based representations. In: INFOVIS, pp. 17-24. IEEE (2004)
2. Kornaropoulos, E.M., Tollis, I.G.: Overloaded orthogonal drawings. In: Speckmann, B. (ed.)
GD 2011. LNCS, vol. 7034, pp. 242-253. Springer, Heidelberg (2011)
3. Sugiyama, K., Tagawa, S., Toda, M.: Methods for visual understanding of hierarchical system
structures. IEEE Tran. on Sys., Man, and Cyb. 11(2), 109-125 (1981)
4. Tamassia, R.: On embedding a graph in the grid with the minimumnumber of bends. SIAM
J. on Comp. 16(3), 421-444 (1987)
