Some flexible information retrieval systems using structure matching procedures

Abstract

The comparison between stored information identifications and requests for information is one of the principal tasks to be performed in automatic information retrieval. In so-called descriptor systems, where information is represented by sets of independent key words, this operation is relatively simple, since it consists of a comparison between the respective "vertors" of key words. In many retrieval systems it has been found necessary or expedient to use more complicated constructs for the identification of information. Notably "role" indicators are often added to identify various types of key words, and "links" specify a variety of relations between key words. A complete identification for a document or an item of information is often represented by a graph, consisting of nodes and branches between nodes, to identify respectively the key words and relations between key words. The matching of such information graphs with graphs representing requests for information is a relatively complicated and time consuming operation, particularly since the request structure can be made to match the information structure only partially and incompletely.