Information Technology Reference
In-Depth Information
values in the lower row should, therefore, be seen as most important from the
ones displayed.
A careful inspection of the linear regression results associated with parame-
terizing vector
provides additional insights. In particular, we are interested
in the observed
R
2
values and the beta coe
cients (also known as “standardized
coe
cients”). The
R
2
for the whole model set, as well as the average value for
the K-fold cross validation is 0.52. This value shows the explained level of vari-
ation in abstraction style as explained by the various distance measures under
consideration and can be considered as moderately strong. The beta coecients
of the distance measures in various spaces reveal their impact on the activity ag-
gregation. The beta coecients for activity property types
Role
and
Responsible
role
have average values of 0.55 and 0.37, respectively. At the same time, the
standardized coecients of
Neighbor
and
Label
property types fluctuate around
0. The average value for
IT systems
is in between, with a beta coecient of
0.19. The provided numbers illustrate that the activity property types
Role
and
Responsible role
have a big impact on the abstraction style of the considered pro-
cess model collection.
IT systems
also contributes to the activity aggregation,
but the influence of activity labels and activity neighborhood is insignificant.
Clearly, such insights may differ from one process model to the other.
The validation indicates that the suggested distance measures can be used
in a close approximation of the abstraction style of human modelers. Among
the introduced measures,
dist
agg
is of great interest, as it takes into account
the abstraction style of a particular process model collection. Furthermore, the
validation revealed activity property types,
Role
and
Responsible role
,thathave
the highest impact on the abstraction style for this particular collection.
W
4 Related Work
The topic of business process model abstraction can be related to several research
streams. We identify these streams looking both from the perspective of the
disciplines of software engineering and business process management.
Model properties and relations are thoroughly investigated in the software
engineering area. For instance, in [21,22] Kuhne elaborates on the concepts of
model, metamodel, model types, and model relations. These works systematically
describe and organize relations, e.g., generalization and classification, which are
seminal for the problem of model abstraction. Closely related are also the studies
that cover model granularity. In [17], the authors investigate model and meta-
model granularity. The authors compare several metamodels and come up with
best practices with respect to granularity. One can observe that the relation be-
tween a coarse-grained activity in an abstract model and its counterparts in the
initial model is the meronymy, or part-of, relation. Meronymy has been studied
in depth in the software engineering domain [3,13]. Although the referenced pa-
pers do not provide concrete techniques for the implementation of abstraction
within process models, they facilitate a better problem understanding and help
to identify the main concepts in this domain.
