Information Technology Reference
In-Depth Information
As we can see that, processes (a)-(d) are not completely equivalent to one
another. Clearly, in real-life applications, it is not that helpful with a binary
answer (yes/no for the equivalence) when comparing process models. In the next
part, we propose a process similarity metric to determine the degree of similarity
and provide a gold standard alignment for the later experimental evaluation.
2.2 Process Similarity Metric
To compare pairs of process models, we define a metric on the semantic and
topological basis. In this paper, when a pair of process models contains similar
patterns, they should fulfill the following conditions:
Given two graphs G1(N1,E1,F1) and G2(N2, E2,F2), a similar pattern pair
is defined as (W,V), where:
-W=
w
1
w
2
...
wi
{path nodes between wj and wj
+1
}
,wherei=
0,1,2,..,
0
≤ j<i
-V=
v
1
v
2
...
vi
{path nodes between vj and vj
+1
}
,wherei=
0,1,2,..,
0
≤ j<i
- For each pair of correspondence sets(wm, vn), where
0
≤ m
,n
<
the number
of found correspondence sets, we have:
If
m
=
n
, any nodes in wm has low degree of similarity to any nodes in vn
If m = n, the nodes in wm has high degree of similarity to any nodes in vn
Fig. 2.
Standard retrieved pattern from process (a) and (b)
To apply this metric to the models under study in this paper, we select Process
(b) as the target model. Next we manually compare each of the three process
model pairs and come out with a gold standard alignment for the experimental
evaluation part. This manual comparison is done by three students including
one doctor.
Figure 2-4 mark with the gold standard summarized in Table 1. Let's look at
Figure 2. We have correspondence sets: (a1, b1), (a2, b2), (a3, b3) and (a4, b4).
In each correspondence set, any node from the first set is semantic similar to
any node from the second set. These correspondence sets are leveraged to for-
mulate the similar pattern between the two process models. By including nodes
