Ph.D. Dissertation Defense by Nsikan J. Udoyen
Monday, November 29, 2004
(Dr. David Rosen, Chair)
"Information Modeling for Intent Based Characterization and Retrieval of Finite Element Analysis Models"
Archived finite element analysis (FEA) models are a source of information often referred to in the development of new models. Automated retrieval of these models from electronic repositories presents a unique information retrieval scenario because they are supported by information in other formats which must be pieced together to explain its content. These range from spatial representations such as CAD models, sketches and diagrams, to natural language accounts such as notes, reports and annotations in models scripts. Basic search techniques for electronic repositories rely on superficial document attributes and keywords. Such queries are inadequate for open, dynamic repositories, accessible to more users and applications and constantly being expanded with new documents without the restrictions on format that would exist in a database or knowledgebase. The low precision and recall associated with such searches translates into time spent opening files to manually assess their relevance. A more precise means of describing the information sought is required that enables the comparison of intent, content and context to determine relevance.
The aim of the proposed research is to facilitate automated retrieval of relevant
FEA models and related information from open electronic repositories through
the development of formal taxonomies that enable the representation of document
collections as relational data structures. A grammar-based approach with
its foundations in Formal Concept Analysis and the Description Logic ALE
is proposed to enable the creation of these representations, modeling of
the information involved and implementation of queries to find analogous
models. The use of structural subsumption, least common subsumer computation
and concept lattices to assess similarities between representations to support
retrieval of analogous models and supporting documents will be explored.
The impact of their use to incorporate semantics into queries on recall and
query precision will be assessed. Computationally feasible approaches to
implementing these queries will be developed for FEA model retrieval. The
approach will be demonstrated using FEA models of electronic components and
supporting documents stored in electronic repositories.