A THESAURAL MODEL OF INFORMATION-RETRIEVAL

In an information retrieval system both queries and document representatives can be viewed as representations of topics. The central activity of such a system is thus the comparison of one topic representation with another. The set of terms contained in a typical topic representation may be adjusted or extended by reference to a domain thesaurus. This paper proposes that topic representations should actually consist of excerpts from a domain thesaurus, generated by a spreading activation technique. An algorithm for generating excerpts is outlined and exemplified, and the problem of assessing the resemblance between two excerpts is discussed. The paper questions whether existing thesauri would be adequate for this purpose, and offers some ideas on how suitable thesauri might be constructed.