This SGER addresses issues associated with repository development and management. A key issue to be addressed is automatic assessment of the "relevance" of a resource to a topical domain without the mediation of a human domain expert. To achieve this, the project will investigate, develop and evaluate machine learning methods for building relevance models. Previous work by the investigators with models "trained" to automatically judge quality of a resource will be applied. These will be used in conjunction with domain independent quality indicators. Among the general quality indicators explored to date are presence of graphics and explanatory illustrations, good writing style and structure, and authoritative references. The models are evaluated by comparing their performance and results with those of human experts. When performance of the models approximate expert human performance, tools can be produced that can expedite and simplify digital repository management.
