Information Storage and Retrieval最新文献

An experiment in superficial indexing was carried out to elaborate the contrast between indexing with a controlled vocabulary and using automated methods as described by salton [1, 2]. Superficial indexing refers to a set of procedures for selecting, organizing and using indexing terms modelled after automatic techniques but requiring neither the full text of articles, abstracts or machine readable text. It represents a minimal effort in manual indexing designed, in part, to allow the development of an automated version.

Retrieval comparisons were made with a manual system using controlled techniques, and though they were not as favorable as the results reported by Salton, still favor development of such techniques over manual systems.

The purpose of this paper is to report the results of experiments in document clustering using three well known test collections. Automatic classification is briefly introduced. The hypothesis underlying the use of clustering is discussed. A framework for the evaluation of cluster-based retrieval strategies is constructed. These strategies are shown to be dependent on the method of cluster representation (cluster profile) adopted. Finally, a particular cluster-based strategy together with a cluster representation method associated with it is examined and evaluated in detail.

This paper reports on a series of programs that have been developed to process data-bases, consisting of textual items, and to index and arrange (classify) the data items in accordance with an automatically generated classification system. The programs produce directories and a classification-number ordered data-base on microfilm, where it may be searched using a microfilm reader, or on magnetic tape for input to an on-line computer system for search and retrieval.

In the automatic indexing, candidate index words or phrases are selected automatically and the user can reject unsuitable candidate words, en masse, based on various listings prepared by the computer.

The automatic classification of text items in conjunction with the indexing produces a number of useful features. First, it allows the grouping of “like” items on a print-out, on microfilm or in computer storage. This organization of the items is useful for searching and understanding of the content of the data-base. Searches by conjunction of index terms are simplified and can be performed manually (or automatically) using the directories that are produced.

Automatic processing of raw text data, and requiring little user work make the system attractive where low cost is imperative, such as in private or specialised data-bases.

The binary search scheme is a method of finding a particular file from a set of ordered files stored in a computer. This scheme involves examining the middle file, and eliminating those files which do not contain the required file. This procedure is repeated until the required file is found.

In this paper an exact expression for the expected number of passes required to find a file selected at random is derived. The expected number of passes is shown to be asymptotically equal to log₂ k where k is the number of files. The asymptotic approximate values are compared to the exact values.

The term “word government” is used to convey the concept that some words are habitually used with certain constructions, which in a sense they control or govern. A “governing word” is one which requires one or more of several constructions to complete its meaning, and a “word government table” is a tabulation of such words with information about their parts of speech, patterns of use, meanings and meaning-relationships with other words in the construction. Because governing words are the troublesome words of the language—high usage words which have acquired many meanings and usages—tabulating such words and documenting their use pay rich dividends in text processing applications. As part of a project in automatic indexing and abstracting at the Lockheed Palo Alto Research Laboratory, government tables have been compiled for all governing words beginning with the letters A through R ; it is estimated that the complete tables will contain fewer than 8000 governing word entries. This paper first describes how the format of the tables has developed as the tables have been compiled. The use of the tables for resolving syntactic and semantic ambiguities is indicated in some detail. Some practical applications of word government analysis are discussed. Finally, word government techniques are compared to other approaches which deal with the problems of capturing meaning for text processing purposes, and some of the limitations of word government are noted.