Classification Space: A Multivariate Procedure For Automatic? Document Indexing And Retrieval

A conceptual approach to linguistic data processing problems is sketched and empirical illustrations are presented of the major software components- indexing, storage, and retrieval-of a document processing system which offers, in principle, the advantages of complete automation, unlimited cross- indexing, effective sequential retrieval, sub-documentary indexing reflecting heterogeneity of subject matter within a document, and a procedure for automatically identifying retrieval requests which would be inadequately handled by the system. The indexing schema, designated as a "Classification Space" consists of a Euclidean model for mapping subject matter similarity within a given subject matter domain. A schema of this kind is empirically derived for certain fields of Engineering and Chemistry. A set of five related empirical studies provide convincing evidence that when appropriate experimental procedures are followed a very stable C-Space for a given content domain can be constructed on a surprisingly small data base. Other empirical studies demonstrate specific computational procedures for effective automatic indexing of documents in a C-Space, using a relatively small system vocabulary. One study demonstrates that a C-Space maps subject matter relevance as well as subject matter similarity, and thereby pro- motes effective sequential retrieval ; this result is also shown under conditions of automatic indexing. Negative results are found in an attempt to use the structural linguistic distinction of subject and object as a means of improving techniques for automatic indexing.