The K - Rep System

1 Introduction K-Rep is an object-oriented knowledge representation system based on description logic Woods and Schmolze, 1992]. K-Rep provides a language to express descriptions (e.g., of medical concepts such as drugs, treatments, and diseases), inference mechanisms for reasoning about descriptions and their relationship to one another, and considerable supporting infrastructure. This paper presents an overview of K-Rep's system architecture, which has been motivated in part by the needs of several applications in the area of clinical information systems Mays et al., 1996]. It has become clear that accurate and comprehensive controlled medical terminologies are crucial for advanced clinical information systems, e.g., to achieve comparable data across the enterprise in support of outcomes analysis and inter-operation of ancillary systems, conceptual indexing of clinical advice, and knowledge-based entry of orders for drugs, procedures, etc. Organization and maintenance of such a terminology is a formidable task: there are at least 100,000 concepts of initial interest , both national (or international) standardization and local customization of content must coexist, and the terminology will evolve signiicantly over time. The state of the art in medical terminologies is characterized by ad hoc deenitions, manual classiication, little inferen-tial power, and no provision for collaborative development. Therefore, the area of clinical information systems has become one of our key applications. In October 1994, after several years of experience with a Common Lisp implementation of the K-Rep description logic system Mays et al., 1991], we embarked upon a substantial redesign eeort to accomplish the following objectives: Scalability: K-Rep must support extremely large knowledge bases, potentially beyond the limits of virtual storage. ist independent of process termination or system shutdown. Distribution: Knowledge bases must be simultaneously accessible from multiple client systems in a networked environment. Embeddability: K-Rep must ooer terminology services as a seamlessly integrated component of larger application systems. Performance: K-Rep must be able to respond at interactive speeds in critical production environments. Authoring support: K-Rep must provide a professional development environment to facilitate the collaborative knowledge engineering and visualiza-tion activities of domain experts in an intuitive and appealing manner. Consistency maintenance: Revision and reclassii-cation of descriptions is essential, both for interactive knowledge base editing and for run-time infer-encing. Scope: K-Rep must accommodate a wide variety of information within descriptions, including both semantic roles and primitives, and non-semantic information such as bitmaps, synonyms, abbreviations , and codes. These considerations led us naturally to our current design point, illustrated in …