We present ONTOMS2, an efficient and scalable ONTOlogy Management System with an incremental reasoning. ONTOMS2 stores an OWL document and processes OWL-QL and SPARQL queries. Especially, ONTOMS2 supports SPARQL Update queries with an incremental instance reasoning of inverseOf, symmetric and transitive properties.
In order to efficiently manage ontology data, various ontology data management
systems [
ONTOMS efficiently manages large sized OWL data based on RDBMS and
using OWL-QL. It stores OWL data into a class based relational schema to
increase the query processing performance. Figure 1 describes an example of OWL
data stored into relational tables in ONTOMS. Unlike other approaches,
ONTOMS generates a class based relational schema, where one relation is created
for each class. Each class relation contains associated properties as its attributes.
For more details of ONTOMS, please refer [
We denote the new version of ONTOMS where our incremental update strategy and SPARQL processor are applied as ONTOMS2. We designed the architecture of ONTOMS2 as depicted in Figure 2. The core modules of ONTOMS2 are OWL Data Storage Module, Instance Inference Module, SPARQL Module, and OWL-QL Module. To apply our incremental update strategy to ONTOMS, we create SPARQL Module in Query Processing Module and modified Instance Inference Module for the incremental reasoning. OWL Data Storage Module obtains class and property hierarchy information from OWL Reasoner, Pellet, and UID GS1 UID
Prof1 Course UID C1 C2
GraduateStudent degreeFrom degreeFrom_S degreeFrom_E
Professor degreeFrom degreeFrom_S degreeFrom_E
Univ1 6 7 GraduateStudent_ Professor_ takesCourse teachesCourse UID Value UID Value GS1 C1 Prof1 C1 GS1 C2 Prof1 C2
OWL
Tuples for Class, Property, Instance
OWL Data
Query Query Result
IInMnfseotradenunclceee IRnecarseomneinntgal SMPAodRuQleLModuOleMWoLd-uQleL
QSuQerLy iRQneuTseuurlyptles
generates relationship information among classes and properties when parsing
the given OWL data. Once OWL data is stored, Instance Inference Module
performs instance reasoning with properties which are collected along with their
values from RDBMS, and stores newly generated instances into RDBMS. The
details of instance reasoning for initial OWL data can be found at Section 6
in [
SPARQL Update There are several query languages for OWL such as
OWLQL and SQWRL . However, all of them support a read-only(select) query only.
The previous version [
According to the SPARQL Update recommendation proposed by W3C, SPARQL
Update provides three operations related to update : INSERT DATA, DELETE
DATA, and DELETE/INSERT. The INSERT DATA operation is to add new
triples into the ontology data while the DELETE DATA operation is to remove
triples from the ontology data. Lastly, the DELETE/INSERT operation is to
remove triples and add new triples into the ontology data with the WHERE clause.
ONTOMS2 supports all INSERT DATA, DELETE DATA, and DELETE/INSERT
operations. Due to the space limitation, we omitted the detailed syntax. Please
refer Chapter 3 SPARQL 1.1 Update Language in [
OWL defines several types of properties. However, only inverseOf,
symmetric and transitive properties may generate new facts(triples). The incremental
reasoning for the inverseOf and symmetric properties can be done in a straight
forward way. For the insertion and deletion of transitive property triples, we
adapt the SQL-based transitive closure maintenance algorithm presented in [
JENA insertion ONTOMS insertion JENA deletion ONTOMS deletion 3 80000 70000 s60000 e l ip50000 r td40000 e rer30000 f In20000 10000 0
Inferred triples 0 2000 4000 6000 8000
Insert triples (a) The number of inferred triples 10000 0 2000 4000 6000 8000
Insert / delete triples (b) The update processing time 10000
The most important enhanced point of ONTOMS2 over ONTOMS is that ONTOMS2 supports an efficient ontology data update with an incremental reasoning. Therefore, we focused on the incremental reasoning performance when ontology data is updated. We compare ONTOMS2 with JENA which is one of the most popular ontology data management systems. JENA provides the SPARQL update feature through ARQ. However, ARQ only supports the instance reasoning with non-update(SELECT) queries1. Therefore, we implemented a simple SPARQL update interface for JANA which uses OntModel. Among reasoners in JENA, OW L M EM M ICRO RU LE IN F is used.
Note that, in JENA, the instance reasoning should be re-run from the scratch
for each update query and the update processing time for JENA does not contain
the time for storing the results of the instance reasoning into the relational tables
while ONTOMS2 does the incremental reasoning and the update processing time
for ONTOMS2 contains the storing time. Our experiments were performed on
2.27GHz Intel Xeon with 12GB of main memory. We implemented ONTOMS2
using MS SQL Server 2008 and Java. For JENA [
Figure 3(a) shows the number of triples to be inferred for various sizes of insertion triples(facts). All fact triples have same transitive property. New triples to be inserted as a facts are generated such that their subjects and objects are randomly selected from subjects and objects of previously inserted fact triples. 1 ARQ supports a SPARQL update query on basic Model only, not on OntModel or
InfModel. However, JENA uses OntModel or InfModel for the reasoning. In JENA, the instance reasoning for each insertion is conducted from the scratch while the instance reasoning of ONTOMS2 is incrementally conducted. As a result, ONTOMS2 outperforms JENA for the insertion and deletion due to the incremental update strategy. The results are as shown in Figure 3(b). 4
During the demonstration, we will illustrate how our system handles SPARQL queries over ontology data such as retrieving instances of classes or properties, inserting/removing instances which satisfy a specific condition with an incremental reasoning. Our system also contains a GUI-based wizard to help a user to create a SPARQL query easily. The screen shots of ONTOMS2 are depicted in Figure 4 and the recorded video can be found at our demo page (http://islab.kaist.ac.kr/ONTOMS2/).
Acknowledgments This work was supported by the National Research Foundation of Korea grant funded by the Korean government (MSIP) (No. NRF2009-0081365).