Automatic Indexing in Information Retrieval

Abstract

This thesis examines some of the problems involved in the implementation of a fully automatic information retrieval system. Semi-automatic systems are already in existence but the difficulty with such systems is the intellectual effort required to index documents. An attempt at automatic indexing is made in this thesis by indexing documents with the terms that are contained in their texts. Documents are then retrieved as relevant to a request, if there are terms in a document that are also in the request. Assessment of the relevance of the documents provides the data from which performance of the system is measured. In an effort to increase the number of relevant documents retrieved, alternative terms to those specifically contained in the request are generated. The performance of such a system is compared with the performance of the system using single terms. The body of information, used to test the systems, is restricted to the subject field of numerical analysis.