Reconstructing conceptual Structures of Relational Databases in Frames Based on Data Dependencies

Abstract

This paper presents an overview of the TRICS system which discovers a conceptual structure of an existing relational database, and its mechanisms for eliciting the conceptual structure of the database from its relational instances. In order to represent the conceptual structure of the database, it is necessary to make conceptual objects and relationships between them clear. Conceptual objects and their relationships are obtained based on functional and inclusion dependencies. These dependences are inductively inferred by analyzing a collection of relational instances. Functional dependencies are used for making conceptual objects clear, while inclusion dependencies for relating these conceptual objects to each other. A conceptual object is specified with a relation which is in third normal form (3NF). If an existing relation is not in 3NF, the relation is decomposed into some 3NF relations. Each conceptual object is transformed into a frame. There are three types of relationships between two frames, they are is-a, is-part-of is-associated-with. A hierarchical structure of frames is constructed based on is-a links. Part-whole relationships are represented in terms of is-part-of links. And, relationships for relating associative objects are represented in is-associated-with links. By eliciting conceptual objects and their relationships, and describing them in terms of frames, a structural representation of a existing relational database is inductively obtained. It will improve the transparency and understandability of the database's structure. Such a structural representation is also useful for integrating the existing database into a knowledgebase system, since the knowledgebase system has to retain a description of the data structure of the database.