Information Technology Reference
In-Depth Information
5
Related Work
Compliance checking of business process models with a focus on execution order con-
straints has been approached from two angles: namely compliance by design and com-
pliance checking of existing models. The latter has been tackled using model checking
techniques [6,10,14]. Our work follows a compliance by design approach that has also
been advocated in [11,12,13,15,16,25]. Close to our work, the authors of [12,11] em-
ploy temporal deontic assignments to specify what can or must be done at a certain
point in time and synthesize a process template from these assignments. In contrast to
our work, however, the approach is limited to temporal dependencies between activity
executions and the underlying logic requires an encoding of these dependencies via
explicit points in time. Another approach to synthesize compliant processes was intro-
duced in [25]. The authors employ a set of compliance patterns expressed in Linear
Temporal Logic (LTL). For each pattern a finite state automaton (FSA) is defined. To
synthesize a process, the FSAs of the involved patterns are composed. Next, the user
is required to select for each composition an execution path in order to synthesize the
process. That approach is able to generate processes with sequence and choice only.
Moreover, it does not consider data flow aspects in the synthesized process.
Related to our approach to process model synthesis is work on process mining, which
aims at automatic construction of a process model from a set of logs [5,4,3]. We adapted
the α -algorithm [4], a standard mining approach, for our purposes. Besides the com-
monalities, there are some important differences between process mining and process
template synthesis. We consider control flow routing based on data values. This aspect
is often neglect in process mining algorithms. Only recently, time information and data
context have been considered when predicting the continuation of a trace based on its
current state [21,2]. Further, process mining approaches have to be robust against incor-
rect data (log noise). As we derive a model from artificially generated traces, this is not
an issue for our approach.
Work on declarative business process modeling is also related to our work. The au-
thors of [17,19] propose to model processes by specifying a set of execution ordering
constraints on a set of activities. These constraints are mapped onto LTL formulas;
which are used to generate an automaton that is used to both guide the execution and
monitor it. That is similar to our approach of generating a pseudomodel. Recently, the
authors also showed how finite traces that respect interleaving semantics can be ex-
tracted from a set of LTL constraints [18]. The major difference from our work is
that [18] does not model data constraints as we do. They also change the semantics
of LTL rather than by using standard LTL as we do. Finally, we initially tried the ap-
proach of extracting Buchi automata from our LTL specifications for our example, but
found that the automata approach required hours to return the automata whereas our
LTL satisfiability checker returns a pseudomodel in less than a second.
6Con lu ion
In this paper, we introduced an approach to synthesize business process templates out of
a set of compliance rules expressed in LTL. We also showed that extra domain-specific
 
Search WWH ::




Custom Search