Dissertation Information for Geoffrey Zhengfu Liu NAME: - Geoffrey Zhengfu Liu
DEGREE:
- Ph.D.
DISCIPLINE:
- Library and Information Science
SCHOOL:
- University of Hawaii (USA) (1994)
ADVISORS: - Larry N. Osborne
COMMITTEE MEMBERS: - David Chin - David Stampe - Laku Chidambaram - Gerald Wayne Lundeen
MPACT Status: Fully Complete
Title: An experimental study of incorporating thematic role analysis into VSM-based text information representation and retrieval: Design and evaluation of the semantic vector space model
Abstract: One of the current trends in the field of information retrieval is to apply artificial intelligence techniques, especially natural language processing and knowledge representation techniques, to the problem of information retrieval. Although this approach is appealing, it is unlikely that the problem can be solved once and for all by completely relying on the semantic processing and knowledge representation techniques and attempting direct retrieval of information from a knowledge base constructed out of a collection of natural language texts.
A feasible approach is to use the well-developed IR techniques as the backbone and incorporate some of the NLP techniques to increase the power of content representation without involving sophisticated processes of semantic interpretation and knowledge representation.
In this dissertation research, a text representation and searching technique, called "the Semantic Vector Space Model" (SVSM), was developed by combining Salton's Vector Space Model (VSM) with heuristic syntax parsing and distributed representation of semantic case structures. In this model, both documents and queries are represented as semantic matrices. A search mechanism was designed to compute the similarity between two semantic matrices and the similarity value was interpreted as the predictor of relevancy.
A prototype system was built to implement this model by modifying the SMART system and using the Xerox P-O-S tagger as the pre-processor of the indexing process. The prototype system, called "SMART++", was used in a series of experiments designed to evaluate the proposed text representation and searching technique in terms of precision, recall, and effectiveness of relevance ranking. The original SMART system was used as the benchmark. Three experimental collections acquired from Cornell University were used in the experiments.
The results of these experiments showed that if documents and queries were too short (typically less than 2 lines in length) our technique was less effective than the Vector Space Model. But with longer documents and queries, especially when original documents were used as queries, we found that the system based on our technique had significantly better performance than the SMART system. This suggests that a significant improvement of system performance can be achieved by combining semantic case structure information with the weighted term representation of texts in a situation where longer queries are available.
| |
MPACT Scores for Geoffrey Zhengfu Liu A = 0
C = 0
A+C = 0
T = 0
G = 0
W = 0
TD = 0
TA = 0
calculated 2008-06-01 21:09:50
Advisors and Advisees Graph
generating graph, please reload |