A multiagent architecture for semantic access to legacy relational databases

Abstract

This paper presents a novel approach to accessing information stored in legacy relational databases (RDB), based on Semantic Web (SW) and multiagent systems (MAS) technologies. Its purpose is to provide the users of enterprise decision-support systems with direct, flexible, and customized access to information, through high-level semantic queries, without the need to modify the underlying legacy databases. This is effected through a distributed architecture, based on agent-oriented intelligent middleware, consisting of servers that contain databases and clients that provide customized access to users. In a server, the meaning of the database structure is captured in the reference ontology, which is exported to clients. In a client, the meaning of concepts specific to a user is captured in the custom ontology of that user, as an overlay that relies on the imported reference ontologies. These system ontologies are built gradually, within the system itself, in a process that automatically generates the basic structures from RDB schemas and incrementally enhances them to full ontologies through human-agent interactions. The approach relies on agents endowed with meta-ontology that assist human actors in the ontology-building process. A part of their role is to find, identify, reference, import, display, and apply relevant knowledge available on the Semantic Web. The development of ontologies in turn permits further delegation of operational tasks to agents. The approach is expected to become increasingly effective with the advancement of the SW infrastructure. The paper explains the conceptual basis of the approach, outlines the distributed system architecture, and focusses on the server functionality, including the development of reference ontology.