On the number of categories for classification

doi:10.1016/0020-0271(69)90002-3

Information Storage and Retrieval

Volume 5, Issue 1, June 1969, Pages 1-6

https://doi.org/10.1016/0020-0271(69)90002-3 Get rights and content

Abstract

It is shown that under certain restrictions, n items should be classified into √n categories if the amount of information (in the Shannon sense) needed to retrieve them is to be a minimum.

References (6)

F.B. Baker
Latent class analysis as an associated model for information retrieval
H. Borko
Measuring the reliability of subject classification by men and machines
Am. Docum.
(1964)
H. Borko et al.
Automatic document classification
J. Ass. comput. Mach.
(1963)

There are more references available in the full text version of this article.

Cited by (2)

A multiple-objective planning methodology for information service managers
1979, Information Processing and Management
The problem of allocation of scarce resources in an organization is frequently complicated by the presence of multiple, often conflicting, managerial objectives and a high degree of ambiguity in the definition of the organization's purpose. This is particularly true within the operating environment of many public and privately operated information centers and special libraries. In recent years, the managers of such centers have been pressed by both the funders and the users of their services to improve the efficiency of their operations. Although techniques are available for determining and evaluating various measures of the performance of these centers, relatively little has been accomplished with respect to the more fundamental decision making problem of allocating resources in order to optimize the achievement level of the various objectives of the decision maker. The problem addressed in this research is the extension of existing techniques to more adequately deal with the resource allocation problem in an ambiguous environment. The technique known as goal programming is extended and a new methodology, termed goal-range programming, is presented. The goal-range programming methodology is related to the information center planning environment through examples derived from the interaction of the authors with several centers.
Automatic indexing and generation of classification systems by algorithm
1973, Information Storage and Retrieval
A system of automatic indexing based on Baye's theorem is described briefly. In assigning 124 documents to 9 categories, there were 97 cases of agreement with professional indexers. Using a correction factor, based on 87 per cent human consistency from other sources, the computer appears then to index with 90 per cent accuracy in this case.
The technique is then used with two randomized sample document groups drawn from nine categories. Each group in turn is used as the basis for indexing the other. The computer knows only the number of categories. After 8 cycles the computer is found to have formed 9 groups consisting of about 50 per cent of documents that were also lumped together by professional indexers on the basis of subject content.
A new measure of performance is proposed and some other applications of the technique indicated.

View full text

On the number of categories for classification

Abstract

Latent class analysis as an associated model for information retrieval

Measuring the reliability of subject classification by men and machines

Am. Docum.

Automatic document classification

J. Ass. comput. Mach.