Advanced Search
Japanese | English

naistar (NAIST Academic Repository) >
学術リポジトリ naistar / NAIST Academic Repository naistar >
テクニカルレポート / Technical Report >

Please use this identifier to cite or link to this item:

Title: Flexible category structure for supporting document retrieval and its evaluation
Authors: Nakagawa, Kokoro
Takata, Yoshiaki
Seki, Hiroyuki
ナカガワ, ココロ
タカタ, ヨシアキ
セキ, ヒロユキ
仲川, こころ
高田, 喜朗
関, 浩之
Issue Date: Dec-2000
Publisher: Nara Institute of Science and Technology
Series/Report no.: Information Science Technical Report ~ TR2000013
Abstract: A method for supporting document retrieval by constructing a flexible category structure is proposed. In this method, a category structure suitable for retrieval by the user is constructed whenever a query is submitted. The method uses categorization viewpoints as a priori knowledge, where a categorization viewpoint is a finite set of category names. A set of documents retrieved by initial keywords is decomposed by categoriza-tion viewpoints and each decomposition is scored by clearness or entropy. First, the system presents high-scored decompositions, and then the user selects an appropriate decomposition by considering the score. The decomposition can be recursively per-formed until a category structure of suitable size is obtained. By using the experimental system which has 68 categorization viewpoints, the pro-posed method was evaluated with BMIR-J2 test collection. Preliminary experiments show that the set of documents decomposed by the proposed method have higher pre-cision than those decomposed by clustering (K-means). An experiment using human subjects was also performed and we obtained the following results: (1) Quality: The proposed method provides better BMIR-match than both the simple keyword-based method and the clustering method, where BMIR-match is the relative number of doc-uments which BMIR-J2 assumes to be relevant to the topic in the set of documents which the subjects decide to be relevant. (2) Efficiency: Using the proposed method, the subjects can find the same number of relevant documents in shorter time than using the other methods. (3) Usability: The answers given by the subjects to the question-naire suggest that the proposed method parallels the keyword-based method in usability even for the users familiar with the keyword-based method.
ISSN: 0919-9527
Text Version: author
Appears in Collections:テクニカルレポート / Technical Report

Files in This Item:

There are no files associated with this item.

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.


Copyright (c) 2007-2012 Nara Institute of Science and Technology All Rights Reserved.
DSpace Software Copyright © 2002-2010  Duraspace - Feedback