Information Technology Reference
In-Depth Information
Knowledge Base
SWRL
Rules
Tropos SI*
Ontology
Rules
Engine
Facts +
Rules
Translator
SI* to OWL
SI* Model
in OWL
SI* model
Model to
OWL
Pattern in
SQWRL
Result Set
Model
Translator
SI* to OWL Query
Pattern
Context in
SI*
SI* --> SQWRL
SI* --> SPARQL
Pattern in
SPARQL
DL Reasoner
KB'
Fig. 6.
System Architecture for Pattern Matching
the problem. The resultset will contain several sets when a pattern matches to several
parts of the problem. Moreover, each set will provide a mapping from a pattern's con-
structs to the problem. We have implemented this approach using Java Platform v1.6
along some features from Protege libraries.
5
Experimental Results
Design of experiments.
To evaluate this approach and its implementation, we have
conducted experiments using a laptop Intel Core2 Duo T7300 2.0GHz, 2Gb DDR2 667.
Through these experiments we intend to assess the performance of our implementation,
and investigate how performance (i.e., execution time) is affected by an increase in
problem size. To make the experiment realistic, we consider the ATM Scenario [17] as
the problem-at-hand. First, the SI* model of the problem is translated to a corresponding
model in OWL-
. Similarly, we translate SERENITY S&D patterns (21 patterns),
defined in [6], into OWL-
DL
queries in SPARQL and SQWRL. The model is then
queried using SPARQL and SQWRL queries to find matches to those patterns. Bigger
models are obtained by cloning the OWL-
DL
model of ATM scenario facilitated by
the “deep copy” feature of Protege. Originally, the model of ATM scenario has the size
of 472 elements composed of 83 nodes and 389 relations
4
. Cloning was performed on
the ATM model by cloning nodes and their respective relations. The cloning process
was not linear; as we could not control the number of relations a node participates in.
Seven models were obtained through this process, starting from a model size of 832
(136 nodes) up to the biggest model with size 6203 (941 nodes).
In the experiment, each pattern is matched against 8 different models. To ensure
stability of “execution time”, we perform 20 executions for each pair (pattern, model)
and used the average of each execution time. Moreover, a manual verification has been
performed to validate the correctness of each pattern match.
DL
4
All datasets can be found at
http://disi.unitn.it/
˜
yudis/lib/exe/fetch.
php?media=files:dataset.rar
