Emergent semantics: Towards self-organizing scientific metadata

Abstract. Tasked with designing a metadata management system for a large scientific data repository, we find that the customary database application development procedure exhibits several disadvantages in this environment. Data cannot be accessed until the system is fully designed and implemented, specialized data modeling skills are required to design an appropriate schema, and once designed, such schemas are intolerant of change. We minimize setup and maintenance costs by automating the database design, data load, and data transformation tasks. Data creators are responsible only for extracting data from heterogeneous sources according to a simple RDF-based data model. The system then loads the data into a generic RDBMS schema. Additional grouping structures to support query formulation and processing are discovered by the system or defined by the users via a web interface. Discovered and imposed structures constitute emergent semantics for otherwise disorganized information.