A Reasoning Broker Framework for Protégé
ABSTRACT We develop a reasoning broker framework for the combined use of existing
OWL reasoners assisted
by means for caching of results, scheduling of reasoning tasks and
online-selection of appropriate
reasoners. Here we demonstrate an integration of the first version
of this reasoning broker framework
into the ProtÂ´egÂ´e ontology engineering environment. The demo allows
for the configurable combined
use of various OWL reasoners with features of parallel execution and
runtime-selection of appropriate
- SourceAvailable from: Juergen Bock
Conference Paper: A Reasoning Broker Framework for OWL[Show abstract] [Hide abstract]
ABSTRACT: Semantic applications that utilise OWL ontologies can benefit from a broad range of OWL reasoning systems, which allow for the inference of implicit knowledge from explicitly given facts and axioms. Different OWL reasoners, however, specialise in different reasoning problems for kinds of ontologies, and hence perform differently in certain reasoning scenarios. This paper presents a reasoning broker framework, which connects to different existing reasoning systems and intelligently delegates reasoning requests. The behaviour of the broker is controlled by exchangeable and configurable broker strategies featuring selection and parallelisation of reasoners, centralised caching, simulated anytime reasoning, and various other potential features. A first experiment shows performance improvement for a sequence of queries compared to the use of different single reasoners.5th International Workshop on OWL: Experiences and Directions (OWLED 2009); 10/2009
- [Show abstract] [Hide abstract]
ABSTRACT: Wissen und Informationen wachsen nicht nur stetig in ihrer Menge, sie stellen heute vielmehr eine bedeutende Ressource vieler Unternehmen dar. Der effiziente Zugriff auf Unternehmenswissen, wie etwa Expertisen, Ansprechpartner, Projekt- und Meilensteinpläne etc. kann Unternehmensprozesse vereinfachen und somit zur Zeit- und Kostenreduktion beitragen. Semantische Technologien bieten zahlreiche Möglichkeiten, um Daten mit Hintergrundinformationen zu ihrer Bedeutung anzureichern und sie mit weiteren relevanten Informationseinheiten zu verbinden. Solche semantische Relationen führen nicht nur zu einer effizienteren Suche in größeren Informationsräumen, sondern unterstützen den Benutzer auch bei diversen Prozessen, wie etwa Editierung, Annotation und Verarbeitung von Informationen. Die komplexen semantischen Strukturen stellen neue Herausforderungen an die Visualisierung dar. SemaVis eine am Fraunhofer IGD entwickelte Kerntechnologie stellt eine komponentenbasierte, modulare Architektur zur Semantik-Visualisierung zur Verfügung, die die Anwendbarkeit der Visualisierungen in verschiedenen Anwendungsszenarien unter Nutzung heterogener Daten für heterogene Benutzer erlaubt.Internet Der Dienste, Edited by Lutz Heuser, Wolfgang Wahlster, 01/2011: chapter Basistechnologien für das Internet der Dienste: pages 19-40; Springer, Berlin, Heidelberg, New York.
A Reasoning Broker Framework for Prot´ eg´ e
J¨ urgen Bock, Tuvshintur Tserendorj,
Yongchun Xu, Jens Wissmann and Stephan Grimm
FZI Research Center for Information Technology
at the University of Karlsruhe
We develop a reasoning broker framework for the combined use of existing OWL reasoners assisted
by means for caching of results, scheduling of reasoning tasks and online-selection of appropriate
reasoners. Here we present an integration of the first version of this reasoning broker framework into
the Prot´ eg´ e ontology engineering environment. This allows for a controlled use of various reasoners
within Prot´ eg´ e, supporting configurable strategies for selection and parallelisation.
For the Web Ontology Language OWL, various systems for reasoning are available and being currently
developed. Different systems focus on different aspects of OWL reasoning and are optimised for different
types of reasoning, such as ABox or TBox reasoning. Moreover, there are specific research prototypes
of reasoners designed for restricted OWL language fragments to achieve efficient reasoning by trading
language expressivity for speed. Hence, not every reasoner can equally efficient process all kinds of
reasoning tasks for every ontology. Some but not all of the available OWL reasoners can be readily used
from within Prot´ eg´ e 4 by integration via the OWL API1, however, no combined use of reasoners
is currently supported—a once selected reasoner is used for performing any reasoning within Prot´ eg´ e
This paper presents a reasoning broker system called HERAKLES2and its integration with Prot´ eg´ e 4.
HERAKLES is designed to connect to different reasoners while it is itself acting as a single OWL reasoner.
It redirects reasoning requests to the different reasoners in a way, such that the user can benefit from the
strengths of all underlying external reasoners and their combination.
Prot´ eg´ e users who want to use a reasoner while authoring ontologies can now select HERAKLES as a
reasoner via the user interface, and get a larger variety of underlying reasoners which are invoked intelli-
gently and strategically regarding multiple reasoning requests, in order to deliver results more efficiently.
Reasoning tasks can be processed in parallel by multiple reasoners and assigned to reasoners most suitable
according to the type of reasoning requested, the parameters of the request or the characteristics of the
ontology at hand.
In the following, we will sketch the features and design principles of our reasoning broker framework
and reflect on its integration with Prot´ eg´ e 4.
Our reasoning broker framework HERAKLES implements the reasoner interface provided by the OWL API
and can thus be used as an ordinary OWL reasoner by any semantic application. However, behind the
scenes HERAKLES manages several external remote reasoners, which can again be any reasoner that is
accessible via the OWL API’s reasoner interface. Currently there are connections to the standard rea-
soners Pellet3, FaCT++4, and HermiT5, as well as to KAON26via a newly developed adapter bridging
2The presented research was funded by the German Federal Ministry of Economics (BMWi) under the project THESEUS
Figure 1: Overview of the HERAKLES plug-in
the KAON2 API and the OWL API. Furthermore we provide a connection to the approximate reasoning
systems Screech  and AQA , which are both based on KAON2.
Figure 1 illustrates the general architecture of HERAKLES and its integration with Prot´ eg´ e. Apart
from using HERAKLES as a reasoner via the OWL API, it can also be monitored and configured via
tabs and views added to Prot´ eg´ e. The external remote reasoners are connected to HERAKLES via a
remote interface, and can hence be located on different servers.
The centralised control of various external remote reasoners enables the implementation and com-
bination of different features that provide added value compared to traditional reasoner systems. The
following list of features can be implemented by the use of a strategy concept, which control the behaviour
of the reasoning broker system. However, not all of those features have been implemented yet. Concrete
strategies that are in fact implemented are mentioned at the end of this section.
among the available ones are automatically selected at runtime, taking into account the different specifics
and strengths of a reasoner. Selection can depend on a combination of the type of reasoning task (e.g.
ABox vs. TBox, parameters of a query, etc.), the properties of the ontology (e.g. expressivity, TBox/ABox
ratio, etc.) and the characteristics of the reasoners.
Depending on the reasoning task to be performed, the most suitable reasoners
Parallel Execution of Reasoning Tasks
in which several reasoners are available, possibly running on remote machines. The first reasoner to finish
can then propagate the result to Prot´ eg´ e, which results in a gain of efficiency. This way of parallelisation is
currently implemented, however, we plan to extend this idea to splitting the query and executing subtasks
in parallel. The effect of applying parallelisation also brings a benefit if several requests are to be answerd
simultaneously or at frequent intervals. In that case, the quickest reasoner can immediately handle the
next query, and as soon as other reasoners finish the initial task, they are available for consequent tasks.
A reasoning task can be executed in parallel given a setting
Partitioning of Ontologies
an ontology can be split in multiple partitions, such that for some reasoning tasks an execution on this
partition might be sufficient, which also results in a gain of efficiency by reducing the size of the ontology
to reason with. Query reformulation can then be accomplished similar to recent work by Clark&Parsia7.
By application of module extraction techniques such as described in ,
ing requests can be handled concurrently by scheduling different tasks for different reasoners. The use of
parallelisation and selection also allows for answering a larger number of consecutive requests.
Having a multitude of external remote reasoners in the background, multiple reason-
stantly benchmarking them during execution. These statistics can then be used for future reasoner
selections, in order to predict the most adequate reasoners with higher confidence.
Parallel execution of different external remote reasoners allows for con-
Utilisation of the different external remote reasoners is managed by two kinds of strategies: a load
strategy to manage the loading of ontologies into different external remote reasoners, and an execution
Figure 2: Part of the HERAKLES configuration tab.
strategy to delegate API methods for reasoning tasks. Load and execution strategy are both exchangeable,
such that the behaviour of the reasoning broker can be controlled by using customised strategies.
In the current HERAKLES release we provide several strategies, which realise some, though not yet
all of the features mentioned above.
A BasicLoadStrategy loads a set of ontologies into all available reasoners. A range of execution strate-
gies are available. The BasicParallelisationStrategy executes a reasoning request on all available reasoners
in parallel and returns the result of the reasoner, which finishes first. An ABoxTBoxRatioSelectionStrat-
egy selects a set of reasoners which are supposed to deal best with ontologies that have a certain ratio
of ABox and TBox sizes. This information is part of the properties of each reasoner and is based on
benchmarking results . A TaskSelectionStrategy selects suitable reasoners according to the information
about which reasoners work best for which reasoning tasks. This strategy can be configured in detail
from within Prot´ eg´ e, as will be discussed in Sect. 3. A strategy independent caching mechanism allows
for the efficient handling of frequent reasoner invocations, e.g. by the Prot´ eg´ e UI.
In addition to the OWL API reasoner interface, we provide an interface for anytime reasoning, which
is also implemented by HERAKLES. This allows for an asynchronous communication with the reasoner,
where available results are noticed by a listener continuously without the need to wait for the reasoner to
finish. Currently, an AnytimeStrategy is realised by using a combination of different approximate reasoners
and their properties of delivering results that are guaranteed to be sound but possibly incomplete or vice
3Integration into Prot´ eg´ e
We developed a plug-in8for Prot´ eg´ e 4.
additional functionality. We integrated the HERAKLES reasoning broker using the ReasonerFactory
extension point, which makes it selectable from within Prot´ eg´ e, and hence can be used like any other
OWL API reasoner. We also provide user interface components to configure the broker in terms of
selection of remote reasoners, as well as selection and configuration of broker strategies. This is provided
by a new configuration tab. Additionally the usage of the external remote reasoners can be monitored.
Configurations can be persisted in configuration files for reuse.
Prot´ eg´ e provides several OSGi extension points to include
Selection of External Remote Reasoners
soners, and allows to add or remove additional reasoners that can then be used by the strategies. The
user can see the reasoner properties, such as name of reasoner (Pellet, FaCT++, etc.), supported lan-
guage fragment, soundness/completeness characteristics, etc., which helps to decide whether a particular
reasoner is to be included or not. In a specific scenario, for instance, only reasoners that deliver correct
results might be of interest, while approximate reasoning systems are to be disregarded.
The configuration tab displays all connected remote rea-
and execution strategies to be used by HERAKLES. Strategy configuration is currently realised for the
TaskSelectionStrategy, described in Sect. 2 that allows to assign any reasoner function to a reasoner identi-
fied by a set of properties. The selectable tasks correspond to OWL API methods, such as getIndividuals,
The configuration tab allows to define the conditions under which a reasoner can be selected to
execute these functions, based on a combination of reasoner properties. The user can choose between
The configuration tab also allows for the selection and configuration of load
two different modes to configure the conditions. In basic mode a user just selects the reasoner type,
which should be used for a specific API method. In expert mode selection conditions can be defined
manually using any properties a reasoner needs to match. The property constraints can be defined via
conjunctive or disjunctive combination of conditions. For instance, a user can define the condition, that
for an instance retrieval query via the getIndividuals method only sound and complete reasoners are to
be selected, which are optimised for ABox reasoning. This scenario is illustrated in Fig. 2.
Monitoring of External Remote Reasoners
information about external remote reasoners and to show the status of currently running tasks. As the
reasoners are run competitively, the number of times a reasoner responded first to a given task is displayed
along with the percentage relative to the other reasoners. This information can be useful for developers
of reasoners, as well as for the refinement of the HERAKLES configuration. For instance, if a certain
reasoner performs best in a particular scenario, more instantiations of this reasoner can be set up and
connected to HERAKLES, in order to further increase the overall performance.
We also provide a monitoring component to display
reasoner interface, we also provide an anytime reasoning tab for the successive retrieval of query results.
It extends the Prot´ eg´ e 4 DL Query tab by the feature of timely decoupled displaying of query results
that are incrementally expanded or refined in an asynchronous manner, which allows to present early
results before the underlying reasoning task is completed. In combination with the approximate reasoning
systems mentioned before, we further exploit reasoner properties of soundness and completeness to qualify
query results by means of different colours and strike-through rendering9. In one particular scenario, the
use of an unsound but complete reasoner provides early results for instance retrieval queries, which are
then either confirmed or revised subsequently by a sound but incomplete counter-part run in parallel.
(See  for a description of such scenarios.)
As HERAKLES supports anytime behaviour through the asynchronous anytime
The next steps in the development of HERAKLES include the realisation of the features listed in Sect. 2
that are not yet addressed by existing strategies. More concretely we plan to incorporate modularisa-
tion/partitioning for ontologies and queries, as well as more sophisticated reasoner selection, and load
 J¨ urgen Bock, Peter Haase, Qiu Ji, and Raphael Volz. Benchmarking OWL Reasoners. In Proceedings
of the ARea2008 Workshop, Tenerife, Spain, June 2008.
 Matthew Horridge, Sean Bechhofer, and Olaf Noppens. Igniting the OWL 1.1 Touch Paper: The
OWL API. In Christine Golbreich, Aditya Kalyanpur, and Bijan Parsia, editors, OWLED, volume
258 of CEUR Workshop Proceedings. CEUR-WS.org, 2007.
 Ernesto Jimenez-Ruiz, Bernardo Cuenca Grau, Ulrike Sattler, Thomas Schneider, and Rafael
Berlanga. Safe and economic re-use of ontologies: A logic-based methodology and tool support.
In OWL: Experiences and Directions (OWLED), 2008.
 Sebastian Rudolph, Tuvshintur Tserendorj, and Pascal Hitzler. What Is Approximate Reasoning? In
RR’08: Proceedings of the 2nd International Conference on Web Reasoning and Rule Systems, pages
150–164, Berlin, Heidelberg, 2008. Springer-Verlag.
 Tuvshintur Tserendorj, Stephan Grimm, and Pascal Hitzler.
Technical report, FZI at University of Karlsruhe, Germany, December 2008.
Approximate Instance Retrieval.
availavle at http:
 Tuvshintur Tserendorj, Sebastian Rudolph, Markus Kr¨ otzsch, and Pascal Hitzler. Approximate OWL-
Reasoning with Screech. In RR’08: Proceedings of the 2nd International Conference on Web Reasoning
and Rule Systems, pages 165–180, Berlin, Heidelberg, 2008. Springer-Verlag.
9Results from approximate reasoning systems are possibly incorrect or incomplete, which makes it necessary to confirm
preliminary results by using different colours, or to invalidate them by striking them through.