Login

Publications  •  Project Statistics

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

Dissertation Information for Brook Wu

NAME:
- Brook Wu
- (Alias) Yi-Fang Wu

DEGREE:
- Ph.D.

DISCIPLINE:
- Library and Information Science

SCHOOL:
- SUNY, Albany (USA) (2001)

ADVISORS:
- Jagdish Gangolly

COMMITTEE MEMBERS:
- Peter Bloniarz
- David Andersen
- Hemalata Iyer

MPACT Status: Fully Complete

Title: Automatic concept organization: Organizing concepts from text through probability of co-occurrence analysis (POCA)

Abstract: Organizing concepts from large amounts of text requires a sophisticated technique. Traditional manual methods create products, such as thesauri, that have no word-sense ambiguity, but they have difficulties in keeping up with advances of knowledge development. Automatic methods that are based on term frequencies and co-occurrences, such as clustering and neural networks, in contrast, create collection-dependent products, but they cannot resolve the word-sense ambiguity problem. To develop a better concept organization technique, this study found a solution by combining the strengths of manual and automatic methods and mitigating the weaknesses of both. Based on the "subsumption" rule developed by Sanderson and Croft (1999), the dissertation has developed "Probability of Co-occurrence Analysis" (POCA) for automatic concept organization. Concept hierarchies generated by POCA are intended to provide users with an overview to concepts in a text, and the hierarchical structure of the overview represents frequencies and co-occurrences of concepts.

Two experiments were conducted using two separate corpora to investigate the performance of POCA, and the ASIS Thesaurus of Information Science and Librarianship was used to evaluate the results. The first experiment evaluated POCA alone. The outcomes showed that precision, recall, and term inclusion of term pairs derived varied greatly from one subexperiment to another, due to data sparseness and the use of the thesaurus for evaluation. However, the results found that a lower threshold, 0.8, consistently generated better or the same recall and term inclusion without sacrificing precision. The results also showed that POCA was able to create complete and partial collection-dependent concept hierarchies. The second experiment compared the outcomes produced by POCA and Divisive Analysis clustering. The results showed that POCA performed better; and furthermore, POCA might be able to alleviate the word-sense ambiguity in automatic concept organization.

Based on the results of experiments, the dissertation concludes that in the current stage of development, POCA is not sophisticated enough to produce results comparable to those obtained by manual methods. However, the findings of this study provide a new perspective on the development of automatic concept organization techniques.

MPACT Scores for Brook Wu

A = 0
C = 0
A+C = 0
T = 0
G = 0
W = 0
TD = 0
TA = 0
calculated 2008-01-31 06:13:24

Advisors and Advisees Graph