400  ◾  Jon Espen Ingvaldsen
15.3 Architecture
EVS is a process mining framework developed by Businesscape AS. An overview
of the architecture is shown in Figure15.1. e backbone of the EVS architecture
and its search functionality is a graph database integrated with an Apache Lucene
search engine. e graph database consists of serialized trace networks and the
search engine enables fast retrieval of the traces based on textual queries. Adapters
for importing event data are developed for MXML (and SA-MXML), Excel, and
SAP database tables. e rst version of the SAP adapter is described in Ingvaldsen
and Gulla (2007).
e trace networks in the graph database are structured according to a dened
upper ontology. Figure15.2 shows this upper ontology and three related application
ontologies. e upper ontology aims at dening generic and universally valid con-
cepts that integrate the application ontologies in a BPI context. As in ontology lan-
guages like OWL, every instance and dened class is a member of the thing class.
A trace is a sequence of events ordered by the execution time stamp. In addition
to containing time stamp information, events refer to dened business processes
(optional) and a set of static context entity instances involved in their executions.
Figure 15.1 Overview of EVS architecture.
Ontology-Driven Business Process Intelligence  ◾  401
Context entities describe a set of property values (i.e., value of a purchase order,
expected delivery date of ordered goods) and refer to other context entities by use of
associations or part-of relationships (i.e., a purchase order can be associated with a
vendor, and a user can be part of a group). Business processes, context entities, and
events can be classied as nodes in drill-down hierarchies.
e application ontologies dene concepts specic to the settings of actual BPI
projects. In Figure15.2, we can see that four application ontologies are attached to
the upper ontology. ese ontologies are (1) an event class ontology that describes
the breakdown structures of dened services, (2) an organizational hierarchy, and
Figure 15.2 (a) Figure notation. (b) Upper ontology and three related applica-
tion ontologies.
402  ◾  Jon Espen Ingvaldsen
(3) a small ontology describing VRU outputs. Note that Figure15.2 shows only the
class level of the upper ontology and the application ontologies.
e instance levels of the ontologies appear in Figure15.3. In (a), the part-of rela-
tionships between instances in the organizational hierarchy are described. Wells Bank
consists of two departments that further consist of groups and employees and/or serv-
ers. Figure15.3(b) shows how the data attributes from MXML are interpreted as onto-
logical information where the XML attribute name is used to construct an ontological
class, while the attribute values are considered instances of this class. In our example,
we have a data attribute named VRU Output. is attribute name is used to construct
a unique ontological class. e four output values for this data attribute are considered
as instances of this class. Figure15.3(c) shows a trace that was constructed based on
our MXML example. It contains three events that are instances of three event classes
(VRU Start, VRU Complete, and New Customer Start) and refer to three context
entity instances (VRU Server, NC, and Michael). e trace neighbors have further
relations into the application ontologies and eventually the upper ontology.
All the ontologies and trace structures are merged together in the graph database
of EVS. We will refer to this merged ontology structure as the ontology throughout
this chapter. e traces are indexed and made searchable for users. Textual names
and descriptions in their neighborhood networks are gathered to form a searchable
index row for each trace. For our trace 33139, phrases from the related ontologies
like VRU Start, Michael, Team 4, IT Department, CRM, etc. are included to cre-
ate an indexed and searchable description.
Ingvaldsen and Gulla (2008) describe indexing of traces in an earlier prototype
version of EVS. is prototype used Lucene both for indexing of searchable terms
and for object serialization. In the current version of EVS, the responsibility for
serialization of object structures is handled by the graph database, while the search
engine and its index are responsible for making the object structures searchable.
Both databases and search engines are concerned with similar functions to store,
access, and manage information. In contrast to databases, a search engine is more
loosely organized, with no schema that dene data attributes. e Lucene API uses
eld names to assign text segments to dierent attributes of the indexed items, but
does not restrict the type of text that can be stored in a eld (Konchady 2008).
In addition to the graph database and search index, we have a user interface for
querying traces and three dierent modules for analyzing dierent aspects of the
dataset and highlighted trace clusters.
15.4 Process Analysis Approach
In EVS, the user can either explore the data through custom search queries or by brows-
ing the data through the ontological drill-down hierarchy. ese hierarchies traverse
drill-down relationships in the ontologies. As shown in Figure15.4, both subclass-of
Ontology-Driven Business Process Intelligence  ◾  403
Figure 15.3 (a) Instance level in organizational hierarchy (only a subset of
employees is shown). (b) Instances related to VRU output ontology. (c) Example
of trace and its relations to context entity instances and event class definitions.
404  ◾  Jon Espen Ingvaldsen
and part-of relationships are considered drill-down relationships. By exploring data
through drill-down hierarchies, a user can increase and decrease the level of detail.
e screenshot in Figure15.5 shows how search and ontological browsing are
aligned in the user interface. On the left side of the screenshot, we can recognize
the event log-related ontologies and structures from Figure15.2 and Figure15.3.
Each node in this hierarchy represents a trace cluster, and the number to the right
of the cluster name shows the number of traces the respective clusters contain. For
the node New Customer start, the population size is 57. is means that there are
57 traces with the event class New Customer start in their contexts.
On the right side of the screenshot, we show the main panel where the user
can dene datasets and investigate them through several analysis modules. In
Figure 15.5, the ow analysis module is currently selected and displayed. e
approach for analyzing process data in EVS appears in Figure15.6. e approach
Figure 15.5 Screenshot from EVS.
Sub-class-of
relationship
Part-of
relationship
Drill-down
relationship
Figure 15.4 Definition of drill-down relationships.
..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset