System for managing and refining structural characteristics discovered from databases

Systems that allow automatic knowledge discovery from databases will play an increasingly important role in building/sharing large scale knowledge bases. Although many systems for knowledge discovery in databases have been proposed, few of them have addressed the capabilities of managing and refining the discovered knowledge. In particular, the contents of most databases are ever changing and erroneous data can be a significant problem in real-world databases. Hence, the process of discovering knowledge from databases is a process based on incipient hypothesis generation/evaluation and refinement/management. The paper describes a system named IIBR (Inheritance Inference Based Refinement) for managing and refining structural characteristics discovered from databases. Structural characteristics are a kind of important regularity hidden in databases, and are denoted by regression models for describing three kinds of functional relations: the exact, strong and weak ones. IIBR is one subsystem of the authors' GLS (Global Learning Scheme) discovery system, and can be cooperatively used with other subsystems of GLS such as KOSI (Knowledge Oriented Statistic Inference). By means of IIBR, the structural characteristics discovered by KOSI can be added to a knowledge base as the deductive rules and the sets of data for showing their errors, and can be easily managed and refined according to data change in a database. IIBR is based on inheritance inference and error analysis, as well as the model representation of knowledge, multiple worlds/levels, and metareasoning in the knowledge-based system KAUS. Experience with a prototype of IIBR implemented by KAUS is discussed.