A HYBRID KNOWLEDGE-BASED APPROACH TO INFORMATION-RETRIEVAL

Abstract

From a couple of years ago, extensive research efforts have been devoted to the knowledge-based approach to information retrieval. These are usually based on thesauri that admit a variety of relations between control vocabularies called index terms. However, there have been few systems to offer sufficient functions enough to convey domain dependent knowledge as well as domain independent one in storing and retrieving information. This paper proposes a knowledge-based approach that facilitates not only domain independent knowledge but domain dependent one to be represented in the context of information retrieval in a simple and structured manner. It can be accomplished by separating the store of the knowledge into two boxes: domain independent knowledge Box (Ibox) and domain dependent knowledge Box (Dbox). Ibox is used as a graph-based thesaurus that permits "is-a" relation between index terms. Dbox describes domain-specific information, such as synonym relationship, some term relationships other than "is-e relation, and generation rules for complex compound phrases. Dbox is used as the front-end to Ibox that helps to convert the arbitrary terminology given by an IR user to the index terms implied in Ibox and reflects domain dependent knowledge. The query evaluation procedure QEP retrieves the documents relevant to a query given as Boolean form, interacting with Ibox and Dbox. It, first, retrieves all related documents and takes the ones of highest rank by Rada's algorithm. We show that Rada's algorithm does not raise discontinuity near zero, nor counter-intuitive result in QEP, while it does in general situation.