3 Reusable Components for Knowledge Base and Database Integration 13.1 Middleware for Information Integration

Organisations increasingly need to integrate their database and knowledge-based systems into an enterprise-wide information system. This need applies to both new and legacy database and knowledge-based systems. This paper argues that modern mid-dleware technology, notably Java and CORBA, provides an eeective integration medium, particularly when combined with software agent technology. Deening the components of an enterprise information system as software agents provides a degree of uniformity which facilitates integration; Java and CORBA middleware provide a solid platform on which to implement the agent-based architecture. The paper is illustrated with an example of a medical information system prototype, featuring the integration of a number of SQL databases and a CLIPS knowledge-based system, integrated by lightweight and reusable Java/CORBA components. There is currently considerable interest in using middleware technology to integrate sources of data and knowledge. Some of these sources are legacy systems (pre-existing databases and expert systems), while others are custom-built services speciically designed to operate as components within distributed information architectures. Some of the common scenarios served by such architectures are: extracting data from databases and providing it as the input to knowledge-based systems, which in turn derive new information; extracting data from databases and knowledge from knowledge bases, and combining both to compose a new information source; extracting and transforming data and knowledge into constraint programs, the solution of which yields new information. The \technology push" behind these kinds of integrated application originates in projects such as the Knowledge Sharing EEort (KSE) 4] and the Stanford Mediators work 8] of the early 1990s. In particular, the KSE work has popularised the notion of viewing information sources as software agents, which interoperate using an agent communication protocol such as KQML 1]. KQML supports both the transmission of information (data and knowledge statements) and the coordination and control aspects of a distributed