Ontology-Driven Business Process Intelligence ◾ 399
Every tag in our event log is a semantically dened class in an underlying ontology
(Van Dongen and Van der Aalst 2005; Aalst 2008), and the log data form instances of
these classes. e process instance elements are the core building blocks in MXML for
describing sequences of related events. e name of an event in MXML is audit trail
entry. Such elements gather information about the activity carried out (work ow model
element), event type, time stamp showing when the event occurred, and user (originator)
executing the event. e data tag is used to add information about the event context.
e fraction from the MXML le shows one instance of the process Handle
Customer Call and its rst three events. e events describe the start and completion of
the automated steps in the process and the start of a new customer registration. As we
can see, the rst two events are executed by the VRU server, and the VRU completion
nishes 9 seconds after the VRU starts. e third event is executed by an employee
named Michael. He picks up the phone 1 second after the VRU operation nishes and
he starts serving the customer. e service is a registration of a new customer.
Both the WorkowModelElement and the Originator are annotated with a con-
cept in an external ontology. In our example, the VRU WorkowModelElement is
annotated with the ontological concept VRU found in the CallCentreTasks.wsml
le. e New Customer WorkowModelElement and all originators are annotated
similarly. To each AuditTrailEntry in this example, an attribute is added under a
data tag to provide additional information about the output from the VRU. Based
on automated phone interactions with customers, the VRU decides the service to
which the customer should be forwarded. e VRU Output attribute contains the
value of this routing decision (NC = new customer, ST = stock trading, RS = regular
service, and SE = service in English). When MXML is parsed in EVS, such data
attributes are also treated as categorical and ontological information. Based on the
MXML le and the two ontology les, the process mining algorithms can extract
knowledge about the process ows and relate these ows to dierent ontological
concepts and aggregation levels.
EVS is designed to handle the full complexity of real ERP systems like SAP. In
the database of SAP, resource and document changes are logged only as they are
completed (not scheduled or started). erefore, EVS considers events as atomic
without any state information. In MXML, an event has a type and can proceed
through dierent stages such as schedule, start, and complete. EVS does not distin-
guish between dierent types of events and considers each event as a single-staged
and completed happening with one execution time stamp. As a result, an event
with multiple stages in MXML is converted into multiple events in EVS with one
event for each stage.
Using event logs with only single-stage events produces some implications for
the potential of some process mining algorithms. Multistage events are core build-
ing blocks for process mining algorithms that try to identify and describe parallel-
ism in business ow (Wen et al. 2004). Even without multiple stages in the event
logs, we can still discover the nature of real business ows and extract meaningful
and valuable process models enriched with statistics and contextual information.