Information Technology Reference
In-Depth Information
elements are semantically
related
and, therefore, more appropriate candidates to
be aggregated into the same activity than activities without shared elements.
A number of recent contributions exist that consider
semantic
aspects for
aggregation, e.g., [8,31]. However, their assumptions, e.g., the existence of an ac-
tivity ontology [31], are too strict for generic use. Our approach is based on the
application of the vector space model, an algebraic model popular in information
retrieval [28]. As we will discuss in this paper, the use of vector spaces allows
to determine the degree of similarity between activities according to several in-
formation types available in process models. We have validated the proposed
technique applying it to a process model repository that is in use by a large
European telecommunication organization. The repository incorporates hierar-
chical relations between high-level activities and the activities that they aggre-
gate. Also, the process models contain various types of semantic information.
The validation suggests that our approach closely approximates the decisions of
the involved modelers to cluster activities.
The main contribution of this paper is a technique that may assist novice
process modelers in the abstraction of complex process models by mimicking
the abstraction decisions of more experienced modelers, as discovered from ex-
isting models. In this way, the technique allows to reuse activity aggregation
principles for future aggregation decisions. Since the lack of experienced process
modelers is a noted issue in many large modeling projects [26], this is a valuable
asset to improve the process model quality. Meanwhile, the designed technique
can also support experienced modelers enabling process model abstraction in
conformance to their specific abstraction style. Hence, experts can accelerate
their modeling routine configuring this technique, while staying in control over
the modeling outcome. Finally, the technique can also be used to safeguard a
particular “fingerprint” of a process model collection with respect to abstraction
choices.
The paper is structured accordingly. We continue in Section 2 explaining the
proposed algorithm, along with providing the required background knowledge.
Section 3 empirically validates the proposed approach, using an industrial set of
process models from the telecommunication sector. Finally, Section 4 contrasts
our contribution with the related research, while Section 5 concludes the paper.
2 Activity Aggregation
This section elaborates on the proposed activity aggregation algorithm. After
the introduction of the main concepts, we argue how activity aggregation can be
interpreted as a clustering problem. We discuss a suitable clustering algorithm
and alternative activity distance measures. The section focuses on one specific
measure that enables the tuning of an activity aggregation. We explain how the
aggregation setup is realized and show how the setup information can be mined
from an existing process model collection.
