Configuring Intelligent Mediators Using Ontologies

This paper presents a new intelligent mediators configuration approach which exploits high expressive description logics to represent metadata, and reasoning tasks in order to build more flexible mediation systems. A user specifies a needs expression in terms of (i) an interesting view over a given application domain, (ii) sources preferences and (iii) architectural requirements. A well-adapted mediator, is automatically configured according to these needs through a reasoning-based configuration process. A configured mediator can therefore be adapted in order to build knowledge-based mediation systems with an arbitrary architecture. 1 Context and Motivations Mediation systems [1] were introduced to provide an integrated view over distributed and heterogeneous data sources for accessing them in a transparent way. During these last years, their role has constantly evolved. Several mediation approaches, providing different modeling and implementing solutions, have been proposed. In order to provide query expression and metadata management with more semantics, a particular kind of data integration approach, commonly called knowledge-based mediation system, has been proposed. Differently from classical mediation systems, they use high expressive knowledge representation formalisms, i.e., description logics, as basis for data integration. This allows to have a more precise semantic representation of application domains, and to improve classical mediation tasks with inference capabilities. This work focuses on knowledge-based mediation systems. For this reason, we analyzed many existing approaches according to several aspects such as the integration approach, data model and associated query language, and more particularly the mediation system architecture. According to this latter aspect, knowledge-based mediation systems can be mainly divided into two main categories (cf. Figure 1): – centralized mediation systems [2, 3, 5, 7, 8, 6] are based on a domain ontology acting as an integrated view over a set of distributed and heterogeneous data sources. A user formulates a query over the domain ontology. Then the query is rewritten into a set of local expressions over local sources, which are consequently accessed in a transparent way. The mediator represents the single access point to the system, and local sources are directly accessible from it. – distributed mediation systems [4, 9] aim to integrate a very large number of distributed data sources. This makes the construction of an integrated view over them a very difficult task to achieve. Therefore, query processing becomes a distributed task and the centralized mediator is replaced by a net of cooperative components commonly called peers. Each peer provides a local ontology modeling one or more underlying local sources. A user formulates a query over a peer. If locally retrieved data does not fulfill user expectatives, the query is forwarded to some neighbors peers for execution in order to retrieve mode data. To our knowledge, no mediation system, being able to adapt to both applicative contexts, exists today. Fig. 1. Existing approaches. This paper focuses on the intelligent mediator configuration process of ADEMS and on the role of mediators within a knowledge-based mediation system. The architecture of ADEMS has been previously presented in [10], therefore, the paper gives no details on the mediator internal architecture and on query expression and processing. ADEMS exploits the high expressive description logic SHIQ(D) [12] to represent metadata, and exploits reasoning tasks in order to automatically configure well-adapted mediators. A user specifies a needs expression in terms of (i) the interesting view over a given application domain, (ii) sources preferences and, (iii) architectural requirements. ADEMS configures a well-adapted mediator according to these needs, being able to adapt centralized as well as distributed 1 Neighbor peers are those ones that a peer can directly access. Differently from centralized approach, not all resources are directly accessible from a peer (mediator). 2 ADEMS, an ADaptable and Extensible Mediation Service. architectures. A configured mediator manages metadata as knowledge within a set of ontologies, and exploits inference in order to semantically improve the query processing task. Nevertheless, for a lack of space, this paper mainly focuses on the mediator configuration process, and on the role of mediators within a knowledge-based mediation system. No details on the mediator internal architecture and functions are given. The remainder of this document is organized as follows. Section 2 presents the ADEMS approach. It describes the general architecture of a mediation system, its components, i.e. a set of mediators and sources, and the way they interact. Then it introduces the mediator configuration process. Section 3 illustrates the needs expression structure and shows how its ontological representation allows to better represent the semantics of metadata. Section 4 shows how a needs expression is analyzed and a mediator is configured accordingly, by exploiting reasoning tasks. Metadata involved in such a process is also illustrated. Section 5 discusses on implementation issues and experimental validations. Finally, Section 6 concludes the paper.