Document ranking methods for thesaurus-based boolean retrieval systems

Abstract

Information Retrieval (IR) systems provide users with relevant references satisfying their information need. A major role of IR systems, however, is not just to present relevant references but to help determine which documents are most likely to be relevant to the given requirements. IR systems should provide a sequence of documents which are ranked in decreasing order of query-document similarity. The ranked output makes it possible for users to minimize their time spent to find useful information. Therefore, the document ranking method is an important component of IR systems.

Boolean retrieval systems have been most widely used among commercially available IR systems by reason of efficient retrieval and easy query formulation. When the boolean retrieval system uses the thesaurus as indexing vocabularies, it has additional advantages. First, since index terms are selected from the thesaurus, documents on the same topic can be retrieved by the same thesaurus terms regardless of terminology in the documents. Second, the ability to rank documents can be improved by using term dependence information from the thesaurus. In this thesis, we investigate document ranking methods which can be used in thesaurusbased boolean retrieval systems.

Particular document ranking methods such as Relevance, R-Distance and K-Distance have been applied to thesaurus-based boolean retrieval systems. Though the methods effectively rank documents in many cases by using term dependencies, they have no effective weighting schemes for queries and decuments and also suffer from inappropriate evaluation of boolean operators. We propose the Knowledge-Based Extended Boolean Model (KB-EBM) incorporating the extended boolean model and the knowledge from the thesaurus. KB-EBM avoids the problems of the foregoing methods, and also provides high quality document ranking by using term dependence information.

It has been argued that the conventional fuzzy set model based on the MIN...