Developing document analysis and data extraction tools for entity modelling

Abstract

The entity-relationship approach to conceptual modelling for database design conventionally begins with the analysis of natural language system specifications to identify entities, attributes, and relationships in preparation for the creation of entity models represented in entity-relationship diagrams. This task of document scanning can be both time-consuming and complex, often requiring linguistic knowledge, subject domain knowledge, judgement and intuition. To help alleviate the burden of this aspect of database design, we present some of our research into the development of tools for analysing natural language specifications and extracting candidate entities, attributes, and relationships. Drawing on research in corpus linguistics and terminology science, our research relies on an examination of patterns of word co-occurrence and the use of ‘linguistic cues’. We indicate how we intend integrating our tools into a CASE environment to support database designers during each stage of their work, from the analysis of system specifications through to code generation.