A model of browsing-based conceptual information retrieval is proposed employing two different types of dictionaries, a global dictionary and a local dictionary. A global dictionary with the authorized terms is utilized to capture the commonly acknowledgeable conceptual relation between a query and a document by replacing their keywords with the dictionary terms. The documents are ranked by the conceptual closeness to a query, and are arranged in the form of a user's personal digital library, or pDL. In a pDL a user can browse the arranged documents based on a suggestion about which documents are worth examining. This suggestion is made by the information in a local dictionary that is organized so as to reflect a user's interest and the association of keywords with the documents. Experiments for testing the retrieval performance of utilizing the two types of dictionaries were also performed using standard test collections.

A method of assisting a user in finding the required documents effectively is proposed. A user being informed which documents are worth examining can browse in a digital library (DL) in a linear fashion. Computational evaluations were carried out, and a DL and its navigator are designed and constructed.

This paper discusses a method of efficient browsing in digital libraries arranged so that similar documents are placed close to each other. The basic idea is that some relevant documents exhibit a high similarity to the query, and some other relevant documents exhibit a high similarity to documents that the user has already judged to be relevant. Based on this idea, a general efficient browsing method is presented in which the system successively learns the results of relevance judgments by the user and suggests the relevance of the document to be examined in the following stage. A control procedure for browsing is derived in which the whole image of the document arrangement is revealed, while repeatedly performing browsing for various queries. The effectiveness of the proposed method was evaluated experimentally through comparison to the conventional method using ranking lists. Considering its use on the Internet, a navigator for browsing has also been designed. © 2004 Wiley Periodicals, Inc. Syst Comp Jpn, 35(4): 36–48, 2004; Published online in Wiley InterScience (http://www.interscience.wiley.com). DOI 10.1002/scj.10080