Ontology Engineering for Intellectual Agents
Mikhail Roshchin PhD Student of Volgograd State Technical University, Russia 400050, Russia, Volgograd, Parkhomenko, 43-149, Tel. +7-902-312-1715
[email protected] This paper is intended to be presented as talk. Keywords: Ontology, Agent-based Computing, Logic-on-Demand, Triple Semantic Model, Semantics Abstract: The aim of this contribution is to present concepts and to propose techniques and a methodical support for agent-based computing using “rich” semantic descriptions of agents. Our approach is based upon a Triple Semantic Model concept for which the principles of Logic-on-Demand are defined. Introduction The major part of today’s software systems usually consists of commodity intelligent agents used by all competitors in the respective business domain. Agent’s activities comprise software components, libraries, web services, aspects, etc. Discovering, integrating, supporting lifecycle of an agent can only be done if there is enough information available about its qualities and behaviour, its intended business logic and required or provided guarantees, its interacting with the environment and runtime requirements. But, there is still an open issue about the proper and adequate way to semantically describe intelligent agents for further reasoning on them. In parallel, the idea of Semantic Web with ontology vision evolved within the realm of the internet. It turns out that Semantic Web techniques and the mechanisms needed for adequately describing and relating intellectual agents are quite similar: hence, we choose to solve the problem of describing agents for automated agent-based computing in an heterogeneous environment by introducing a logical formalism for knowledge representation, together with ontologies which serve as main part of the application and application domain description. Approach Logic-on-Demand Concept: Our approach, based on the concept of Logic-on-Demand (LoD), is supposed to overcome the problems of expressivity and complexity by accommodating the description means of the proposed ontology languages to the varying needs and requirements, in particular with respect to decidability. The main purpose of the LoD concept is to provide an adequate and adaptive way that is based on uniform principles for describing all the notions, relations and rules, the behaviour and anything else that proves necessary during the agent description process. To achieve this, LoD means to define a basic logical formalism that is adequate and tailored to the application domain and to incorporate additional logic formalisms and description techniques with further expressivity as optional features that can be used whenever needed. These additional formalisms share notions and terms with the basic formalism which will be grounded syntactically in OWL and semantically in the description logic (DL). DL is sufficient to define a terminology, hierarchical structures of terms, and definitions of concepts and their properties through terms. However, it is, for instance, not sufficient to describe implication rules, modalities and probabilities which are needed for a proper reasoning about an intelligent agent. Therefore, we propose to enrich DL with further description means.
We add dynamic characteristics with respect to the concrete values of the involved notions and terms. For instance, we express behaviour variability in response to different situations in heterogeneous environments. Therefore, rule-based techniques have to be introduced as an extension to DL. And we apply implication operator to the concepts, their instances, and values, expressed in DL. Regarding behaviour and quality properties, we need to take into account the temporal and probabilistic character of that knowledge for which we need to define modal logic constraints. Thus, the notions of DL are enriched with modalities and probability functions. Triple Semantic Model Concept: The Triple Semantic Model defines a distributed computing model for intellectual agents. It provides mechanisms to distinguish between different entities represented within that model. The model is built up with three levels: the Ontology Level (which is a meta-model level), the Dynamic Annotation Level, and the Annotation Level (both belong to the model level). The ontologies on the Ontology Level are intended to provide a general framework, in most cases based on a specific application domain, to describe any kind of agent. Since ontologies enforce proper definitions of the concepts in the application domain, they also play an essential role in standardising the specifications of agent properties, requirements and interfaces with respect to their domain. Searching for an appropriate agent means that certain concepts and properties have to be checked whether they fit or not to the given query parameters. The ontology mapping mechanisms among heterogeneous domains help to identify a match of query parameters and agent properties. The relation of the notions is defined among agent dynamic properties on the ontology level, e.g. run-time scenarios, or behaviour patterns. For these definitions, DL with the extensions, as presented in the principles of our Logic-on-Demand concept, is used. We also need to specify dependencies on actual, dynamically changing circumstances, which have an important influence on the reasoning about certain agent. For instance, we define the notions and rules of cost, trust, and belief, which are dependent on concrete interaction and situation. They are specified on the Dynamic Annotation Level, based on the concept of LoD. Dynamic annotations play the role of mediators between the ontology and the static semantic annotations that describes the agent properties, and in particular its requirements with respect to certain environment characteristics. The Annotation Level comprises the static descriptions of the properties and qualities of agents, which can be used for reasoning. The annotation is expressed in pure DL. Conclusion Our approach proposes the Triple Semantic Model based on principles of Logic-onDemand as the pragmatic and adequate solution for semantic modelling of intelligent agents. It structures the description process and introduces flexibility with respect to the description means. The application of the proposed concepts to real scenarios proves feasibility and shows well selected trade-off between expressivity and complexity. References 1. Peter Graubmann, Evelyn Pfeuffer, Mikhail Roshchin : Web Services Annotation and Reasoning. Position paper, W3C Workshop on Frameworks for Semantics in Web Services, Innsbruck, June 2005. 2. Peter Graubmann, Mikhail Roshchin : Semantically Annotated Software Components. Submitted to 5th Int. Workshop on Software Composition (SC 2006), Vienna, March 2006