work with respect to previous research. In Section 3
we present an extended model for representing
process risks based on the process descriptor notion,
presented first in the work of Lincoln et al (2007),
and extended in this work for the field of new risk
design. Then, we describe our method for designing
new risks in Section 4. Section 5 introduces our
empirical analysis. We conclude in Section 6.
2 RELATED WORK
Most of the efforts invested in developing methods
and tools for designing process models focus on
supporting the design of alternative process steps
within existing process models. Such a method is
presented by Schonenberg et al (2008) aiming to
provide next-activity suggestions during execution
based on historical executions and optimization
goals. Similarly, Gschwind et al (2008) suggest an
approach for helping business users in understanding
the context and consequences of applying pre-
defined patterns during a new process design.
Few works were devoted to the design of brand
new process models within specific and predefined
domains. The work presented by Muller et al (2007)
utilizes the information about a product and its
structure for modeling large process structures.
Reijers et al (2003) present a method, for designing
new manufacturing related processes based on
product specification and required design criteria.
Works in the domain of risk design also focus on
specific risk domains, such as credit risks (Giesecke,
2004; Galindo and Tamayo, 2000), inventory
management risks (Michalski, 2009), and financial
risks (Barbaro and Bagajewicz, 2004). Our work
offers a generic design method that is domain
agnostic and does not rely on product design data. In
addition our work assists in the design of risks rather
than process activities.
A requirement for the support of business
process design involves the performance of a
structured reuse of existing building blocks and pre-
defined patterns that provide context and sequences
(Gschwind et al, 2008). The identification and
choice of relevant process components are widely
based on the analysis of linguistic components -
actions and objects that describe business activities.
Most existing languages for business process
modeling and implementation are activity-centric,
representing processes as a set of activities
connected by control-flow elements indicating the
order of activity execution (Wahler and Kuster,
2008). In recent years, an alternative approach has
been proposed, which is based on objects (or
artifacts/entities/documents) as a central component
for business process modeling and implementation.
Our work supports this approach and focuses on
objects for the purpose of risk identification and
modeling.
Finally, the work of Lincoln et al (2007) presents
the concept of business process descriptor that
decomposes process names into objects, actions and
qualifiers. In this work we take this model a
significant step forward by extending the framework
to support also the representation of risks using a
new taxonomy - the “risk descriptor.”
3 THE DESCRIPTOR MODEL
In the Process Descriptor Catalog model (“PDC”)
(Lincoln et al, 2007) each activity is composed of
one action, one object that the action acts upon, and
possibly one or more action and object qualifiers, as
illustrated in Figure 1, using UML relationship
symbols. Qualifiers provide an additional
description to actions and objects. In particular, a
qualifier of an object is roughly related to an object
state. State-of the art Natural Language Processing
(NLP) systems, e.g., the “Stanford Parser” (Stanford
parser, 2016), can be used to automatically
decompose process and activity names into
process/activity descriptors.
Figure 1: The activity decomposition model.
For example, the activity “Manually Calibrate
the Color Machine” generates an activity descriptor
containing the action “calibrate, the action qualifier
“manually, the object “machine and the object
qualifier “color.” In short, this descriptor can be
represented as the tuple
(calibrate,manually,machine,color) - where the
action and its qualifier are followed by the object
and its qualifier. In general, such tuple can be
represented as (A,AQ,O,OQ), where A represents
the action, AQ represents the action qualifier, O
represents the object and OQ represents the object
qualifier.
Natural Language Processing for Risk Identification in Business Process Repositories