Process Modeling and Retrieval in OWL
We evaluated the process retrieval with a KB of 182 different process models.
These process models are based on ten real world processes
(business processes) from the SAP Business Workflow
library (http://help.sap.com), from the Enterprise Services (ES) Workplace
at the SAP Developer Network (SDN) (http://esworkplace.sap.com/sdn) and
from the process library
We used muliplicators in order to create from each basic process model
between 16 and 25 different models.
This muliplicator modifies the processes using the following operators:
(1) Add and remove activities (2) Change of activity order,
(3) Replace activity with sub- or superactivity and
(4) Add and remove flow conditions.
These process models were transformed into OWL.
The process models contain unconditional decisions (70%), conditional decisions (65 %),
split/fork (40 %) and loops (25 %).
For the evaluation we used the Pellet 2.0.0 reasoner in Java 1.6
running on a computer with 1.7 GHz CPU and 1 GB RAM.
The retrieval is evaluated on four knowledge bases with different size.
The first contains 52, the second contains 107,
and the third and fourth contain 182 process models. In the last two KB
the size of the processes (number of activities) is different.
Due to this generation method
the KB contains many similar process models that only differ in some activities
or their ordering.
Depending on the size of the KB
the evaluation consists of 25 until 35 queries.
In the evaluation each query invokes the reasoner without caching.
The queries covering activity ordering, terminology and modality.
The queries for activity order also considered activities within non-sequential
process parts, like in parallel executions or decisions. For terminology, we queried
also for flow conditions and their complement.