Document retrieval system for retrieving a necessary document

Abstract

One or more important sentence structural functional units are extracted from an important portion of each of a plurality of documents, and one or more input sentence structural functional units are extracted from an input sentence. Each of the functional units is a noun phrase, a verb phrase, a pair of nouns connected each other through a preposition or the like. The important sentence structural functional units of each document are collated with the input sentence structural functional units to detect a matching degree, and a similarity level is determined from the matching degree for each document. Thereafter, occurrence frequencies of retrieval words existing in each documents are collated with occurrence frequencies of query words existing in the input sentence to determine an estimation value according to an estimation function for each document, the documents are ranked on condition that one or more documents of a high similarity level are ranked to a higher rank than one or more documents of a low similarity level and a plurality of documents of the same similarity level are ranked in the order of decreasing their estimation values. Accordingly, because the similarity level is determined for each document by considering the functional units, a user can retrieve one or more necessary documents with a high accuracy.