Login

Publications  •  Project Statistics

Glossary  •  Schools  •  Disciplines
People Search: 
   
Title/Abstract Search: 

Dissertation Information for Sung Been Moon

NAME:
- Sung Been Moon

DEGREE:
- Ph.D.

DISCIPLINE:
- Library and Information Science

SCHOOL:
- University of North Carolina, Chapel Hill (USA) (1993)

ADVISORS:
- Robert M. Losee

COMMITTEE MEMBERS:
- Robert Burgin
- Barbara Marie Wildemuth
- Stephanie Haas
- William M. Shaw Jr.

MPACT Status: Fully Complete

Title: Enhancing performance of full-text retrieval systems using relevance feedback

Abstract: The primary purpose of the study is to improve the low precision often found in full-text retrieval systems. In order to enhance the low precision of full-text retrieval while retaining its high recall, relevance feedback mechanisms based on probabilistic retrieval models (binary independence model and two-Poisson independence model) were employed. This study investigates the performance of full-text retrieval systems before and after the incorporation of relevance feedback.

Main effects and interaction effects of three independent variables ((A) Types of Document Representation, (B) Term Distribution Models, and (C) Incorporating Relevance Feedback) were examined using three repeated measures of ANOVA. It was observed that the tests of the main effect on A and C were significant, but the test of the main effect on B was not significant. There was no significant three-way interaction, but there were two-way interactions between the variables. The pattern of the interaction (between A and B) revealed that the retrieval effectiveness of using abstracts and titles deteriorated when moving from the BI model to the TPI model, while the retrieval effectiveness of full-texts improved. The pattern of the interaction (between B and C) indicated that the incorporation of relevance feedback gained more with the TPI model than the BI model.

The analysis of the overall performance (average precision at 0.25,.50, and 0.75 recall levels), averaged over all 83 queries, showed that retrieval using abstracts and titles with the BI model outperformed the others, both with and without relevance feedback.

On the other hand, it was found that retrieval performance of full-texts significantly improved after the incorporation of relevance feedback. In particular, the analysis of the retrieval performance of the seven queries having large numbers of relevant documents, after the incorporation of relevance feedback, revealed that the precision at relatively high recall levels of full-text retrieval exceeded that of retrieval using abstracts and titles.

The findings of this study suggest that the use of abstracts/titles is better for retrieval than full-texts, and that relevance feedback be incorporated in full-text retrieval systems to enhance the low precision.

MPACT Scores for Sung Been Moon

A = 0
C = 0
A+C = 0
T = 0
G = 0
W = 0
TD = 0
TA = 0
calculated 2008-01-31 06:20:20

Advisors and Advisees Graph