The ARD+ method supports the conceptual design of the XTT2-based rules. However, number of limitations of the method have been identied. In this paper a new approach to the conceptual design of rules is proposed. The main idea comes down to the use of the Semantic Web methods and tools to represent ARD+ and overcome the limitations of the original method. In the approach proposed in this paper an OWL ontology capturing the semantics of the ARD+ model is proposed. Such an ontology models the functional dependencies between rule attributes, as well as the history of the design process. At the same time it is more exible that the original method, opening up possibility of integration with other modeling methods and tools, e.g. UML.
are outlined, and the motivation for its extension given. The proposed Design Process Ontology (DPO) is a new approach to the conceptual design of rules. The DPO is introduced in the subsequent section. Possible directions for the future work are given in the nal section. 2
The Attribute Relationship Diagrams (ARD+) method [
ARD is a general method, that tries to capture two features of the design: the attributes, with functional relations between them, and the hierarchical aspect of the process. The second feature is related to the fact that in practice the knowledge engineering process is a gradual renement of concepts and relations.
In Fig. 1 a simple ARD dependency diagram can be observed. It is in fact one of the phases of the benchmark thermostat case study [7] studied in detail in the HeKatE project. The diagram models a simple dependency read as thermostat Temperature depends on Time specication. This is a general statement currently ARD does not model what the specic dependency is, only a simple fact that some dependency exists.
In the following design stage this model can be rened, by specifying Time as a compound attribute, and later on discovering that the set of newly introduced attributes (Date, Hour, season, and operation ) can be in fact decomposed into two subsets that depend on each other. The nodes of the ARD diagram correspond to so-called characteristics (properties) that are described by one or more attributes. Attributes can be conceptual (general), and physical (specic).
Two transformations of the model are possible: nalization and split. The specication transformation (between Time and Date, Hour, season, and operation) is called nalization , whereas the other one is called split. These are captured in the Transformation Process History diagram (TPH). Together with the ARD dependency diagram they form the ARD Model. In the model on the right (Fig. 3) the black edges correspond to nalization and split transformations, and the blue edges show the functional dependencies.
In general, ARD could be used support the design of both forward and backward chaining rules. However, so far it’s been mainly used for forward chaining. The basic idea is that having the most detailed, specic ARD dependency diagram, rule prototypes can be automatically built. A rule prototype is a pair of sets of attributes present in the rule premise, and a set of decision attributes. The prototype is aimed at an attributive rule language [3], such as XTT 2 [5,6].
Time
Temperature Date Hour season operation Date Hour Date Hour season operation season operation
Temperature
Temperature
Temperature
The ARD+ method [4] is an extension of the original ARD [
The rst problem concerns support for knowledge acquisition in general, and has been addressed in [9]. A practical approach to a partial automation of the attribute and dependency identication process by the use of knowledge discovery methods has been introduced there.
The focus of this paper is to propose a single coherent solution for the three remaining problems. A richer knowledge representation model is proposed. In particular it should: 1. allow to specify dierent classes of functional dependencies, 2. provide more expressive means for the history description, 3. allow to build a single coherent model combining both functional dependencies and history information in a single model.
In the next section a proposal of using standard Semantic Web methods meeting the above mentioned requirements is put forward. 4
The basic idea presented here comes down to proposing an ontology called the Design Process Ontology (DPO) capturing the functional dependencies present in the main ARD diagram and history information captured in the TPH.
In general, an ontology is a knowledge representation [10] that serves as a formal denition of objects in the universe of discourse and relationships among them. The domain is described by means of concepts (classes), roles (properties) and instances (individuals). Relations can be specied both among classes and individuals. Ontologies allow for a formal denition of the vocabulary in the universe of discourse, together with its intended meaning and constraints present in the domain. In this paper the Web Ontology Language (OWL) [11], specically the OWL DL dialect, based on Description Logics (DL) [12] is used.
Similarities of the ontology-based modelling approach and the ARD method has been investigated in [13,14]. Alternative approaches to a mapping between ontology concepts and system attributes have been considered. One of them consists in representing system attributes and characteristics as concepts in an ontology. Another proposal is to treat attributes as instances of a generic class Attribute. Both approaches allow for describing the relations between system attributes using ontology properties. In this paper an ontology based on the former approach is presented.
Design Process Ontology is a proposal of a task ontology [15]. Its aim is to capture the system characteristics together with dependencies among them, as well as represent the gradual renement of the design process. Basically, DPO consists of a general class Attribute and four properties: dependsOn, transformedInto , splitInto, and finalizedInto .
The property dependsOn is very general and may be further specialized. It is used to represent functional dependencies among the system characteristics. At this stage we do not formally specify the semantics, which intuitively may be put as one attribute depends on the other. Functional in this context have a dierent meaning than functional properties used in OWL (owl:FunctionalProperty ), where they denote that the property has an unique value for each instance.
The other three properties ( transformedInto , splitInto and finalizedInto ) denote the TPH relations the design process transformations. A hierarchy of the TPH properties may be introduced as follows (DL convention): split_into v transf ormed_into, f inalized_into v transf ormed_into. The domain and the range of all the properties is the general class Attribute.
DPO may be specialized by concrete ontologies for specic design tasks. In this case system characteristics (conceptual and physical attributes) subclass the Attribute class. All the characteristics and attributes identied in a system are represented as independent classes. The properties may be specialized accordingly, so that they range over concrete system classes rather than the general Attribute class. An example of such an ontology for the Thermostat system is depicted in Fig. 4. The ontology has been built in OWL using ProtegØ.
ARD is a method used in a gradual renement process. As the process progresses, the functional dependencies change and new TPH relations are added. It is worth emphasizing that the historical TPH relations remain unchanged, whereas the functional ones are dierent at each process stage (observe Fig. 2). Thus, the ontology is dierent at various design stages. At a given moment an ontology represents all of the characteristics and attributes identied in the system from the beginning of the design process. All the TPH relations, such as split and nalization are shown. As for the functional dependencies, only the most specic relations identied at certain moment are shown (observe Fig. 4).
Fig. 4. Simple DPO in OWL designed in ProtegØ
The paper concerns the conceptual prototyping of decision rules. The ARD+ method discussed in a paper provides simple means for capturing functional dependencies between attribute present in rules. Moreover, it allows to capture and represent the evolution of the model, the history of the design. However, it has some limitations addressed in the paper.
In order to solve these problems, it is proposed to use the Semantic Web tools in the system conceptual design phase. The proposal is to capture the system elements and various dependencies among them, using an ontology. In the Design Process Ontology , certain specic relations are dened. The ontology presented in this paper includes only the basic relations and serves as the illustration of the approach. A concrete ontology specializing the DPO is equivalent to ARD model of a system. Moreover, it provides a more coherent model, while allowing to introduce more relations, which would not be possible in the original ARD.
Future work concerns further investigation of the possibilities of using ontologies in the design process. This includes formalizing various ARD+ model features in Description Logics. As for now, the dependencies between the system characteristics are modelled with various OWL properties (roles). It will be considered, if some of those relations can be incorporated into class descriptions. Certain formal descriptions would help to verify relations such as split_into.
The set of various dependencies represented in an ontology will be enlarged. The general functional relation depends_on should be specialized, including the dierentiation between AND/OR dependencies (as used in AND/OR graphs in diagnostic systems). The set of TPH relations may also be enriched.
As the design process progresses, the Design Process Ontology changes. The transformations between subsequent ontologies will be analyzed and their formalization will be proposed. Use of rules on top of the DPO is considered. These rules could be possibly introduced as DLP (DL Programs) or expressed in SWRL.
A future requirement not directly addressed here concerns support for certain annotations present in the UML class diagram. In this case, the reworked method would be closer in semantics to the UML-based design. For more details see [16,17]. The possibilities of integrating OWL with UML using Protege4 is also to be discussed. This approach could provide means to use and integrate both Semantic Web technologies as well as classic software engineering methods to design intelligent systems. 3. Ligƒza, A.: Logical Foundations for Rule-Based Systems. Springer-Verlag, Berlin,
Heidelberg (2006) 4. Nalepa, G.J., Wojnicki, I.: Towards formalization of ARD+ conceptual design and renement method. In Wilson, D.C., Lane, H.C., eds.: FLAIRS-21: Proceedings of the twenty-rst international Florida Articial Intelligence Research Society conference: 1517 may 2008, Coconut Grove, Florida, USA, Menlo Park, California, AAAI Press (2008) 353358 5. Nalepa, G.J., Ligƒza, A.: A graphical tabular model for rule-based logic programming and verication. Systems Science 31(2) (2005) 8995 6. Nalepa, G.J., Ligƒza, A.: Hekate methodology, hybrid engineering of intelligent systems. International Journal of Applied Mathematics and Computer Science (2009) accepted for publication. 7. Negnevitsky, M.: Articial Intelligence. A Guide to Intelligent Systems. Addison
Wesley, Harlow, England; London; New York (2002) ISBN 0-201-71159-1. 8. Nalepa, G.J., Wojnicki, I.: Varda rule design and visualization tool-chain. In Dengel, A.R., et al., eds.: KI 2008: Advances in Articial Intelligence: 31st Annual German Conference on AI, KI 2008: Kaiserslautern, Germany, September 2326, 2008. Volume 5243 of LNAI., Berlin; Heidelberg, Springer Verlag (2008) 395396 9. Atzmueller, M., Nalepa, G.J.: A textual subgroup mining approach for rapid ard+ model capture. In Lane, H.C., Guesgen, H.W., eds.: FLAIRS-22: Proceedings of the twenty-second international Florida Articial Intelligence Research Society conference: 1921 May 2009, Sanibel Island, Florida, USA. (2009) 10. van Harmelen, F.: Applying rule-based anomalies to kads inference structures.
ECAI’96 Workshop on Validation, Verication and Renement of KnowledgeBased Systems (1996) 4146 11. McGuinness, D.L., van Harmelen, F.: Owl web ontology language overview, w3c recommendation 10 february 2004. Technical report, W3C (2004) 12. Baader, F., Calvanese, D., McGuinness, D.L., Nardi, D., Patel-Schneider, P.F., eds.: The Description Logic Handbook: Theory, Implementation, and Applications.
Cambridge University Press (2003) 13. Szostek-Janik, J.: Translations of knowledge representations for rule-based systems.
AGH University of Science and Technology (2008) MSc Thesis. 14. Nalepa, G.J., Furma«ska, W.T.: Proposal of a new rule-based inference scheme for the semantic web applications. In: 1st International Conference on Computational Collective Intelligence - Semantic Web, Social Networks & Multiagent Systems. (2009) To be published. 15. Guarino, N.: Formal ontology and information systems. In: Proceedings of the First International Conference on Formal Ontologies in Information Systems. (1998) 3 15 16. Nalepa, G.J., Kluza, K.: Uml representation proposal for xtt rule design method.
In Nalepa, G.J., Baumeister, J., eds.: 4th Workshop on Knowledge Engineering and Software Engineering (KESE2008) at the 32nd German conference on Articial Intelligence: September 23, 2008, Kaiserslautern, Germany, Kaiserslautern, Germany (2008) 3142 17. Nalepa, G.J.: Xtt rules design and implementation with object-oriented methods.
In Lane, H.C., Guesgen, H.W., eds.: FLAIRS-22: Proceedings of the twenty-second international Florida Articial Intelligence Research Society conference: 1921 May 2009, Sanibel Island, Florida, USA. (2009)