Masters Theses

Date of Award

8-1984

Degree Type

Thesis

Degree Name

Master of Science

Major

Computer Science

Major Professor

K C. O'Kane

Abstract

This thesis presents a study consisting of a detailed analysis and evaluation on different techniques in the field of automatic information storage and retrieval system design (using as ,a database 220 abstracts from articles of the magazine Communications of the Association for Computing Machinery).

The evaluation of the different techniques was based on retrieval effectiveness, and recal1-precision graphs reflecting the average precision values at eleven discrete recall points were used as evaluation measures.

To accomplish the desired study an hypothetical model against which the different systems using the techniques to be evaluated were to be compared was constructed. This model consisted of a set of search requests and ideal responses.

The study was concentrated on techniques used in three different stages of the design of a fully automatic retrieval system, including techniques used to normalize or reduce language variability (suffix "s" and word stem dictionaries), techniques used to assign importance factors to each term in the language in order to reflect its usefulness in representing information content (the inverse document frequency, the signal-noise ratio, the term discrimination value and the logical assignment techniques) and those used in the retrieval mechanism to determine the similarity between the search request and each document of the collection (the cosine correlation coefficient and the Parker-Rhodes-Needham correlation coefficient).

The results from the present study showed that the system combining the use of word stem dictionary as language normalization technique, the inverse document frequency technique for term importance assignment, and the use of the cosine correlation coefficient in the retrieval mechanism exhibited the best retrieval performance for the document collection and set of requests considered.

Download
Files over 3MB may be slow to open. For best results, right-click and select "save as..."

Share

COinS