Dissertation Information for Jing-Jye Yang NAME: - Jing-Jye Yang
DEGREE:
- Ph.D.
DISCIPLINE:
- Library and Information Science
SCHOOL:
- University of Pittsburgh (USA) (1993)
ADVISORS: - Robert Korfhage
COMMITTEE MEMBERS: - Edie Rasmussen - Donald Shirey - Paul Munro - Shi-Kuo Chang
MPACT Status: Fully Complete
Title: Use of genetic algorithms for query improvement in information retrieval based on a vector space model
Abstract: Query improvement based on modifications of query term weights can be viewed, in vector space models, as searching a high-dimensional document space for optimal or near optimal queries. A new search method, based on a class of adaptive procedures, genetic algorithms, is suggested in this dissertation research. The objective of this dissertation is to study the feasibility of applying genetic algorithms to query modification, aiming to improve the retrieval effectiveness. The idea is that based on relevance feedback the algorithm modifies query term weights to search for the optimal or near optimal query which retrieves more relevant documents than its predecessors.
Initial experiments based on two document collections have indicated the feasibility of this idea. The results have shown both query performance improvement throughout the genetic process and query individual convergence in the final generation. During the study for finding the best set of the genetic parameters, we developed a hybrid method where the crossover rate and the mutation rate are modified within the genetic process. This has led to further improvement of the performance, compared with other parameter setting we have used.
A further test of the hybrid algorithm was applied to a standard document collection, the Cranfield database, and a large document collection in the TREC project. The outcomes from the two document collections also have shown both performance improvement and query convergence as shown in the previous two databases. Moreover, the results from the Cranfield database also showed the improvement of retrieval effectiveness measuring in standard precision, compared with both baseline case with no query modification and the outcomes from one of other studies using relevance feedback. The outcomes from TREC have shown the applicability of this genetic algorithm to large document collections where additional relevant documents were brought to users in the genetic modification process.
Several features of the genetic query modification also have been displayed. The algorithm has shown adaptability to different evaluation functions where user's preference for precision versus recall can be taken into account by tuning the system. Another feature is parallel search where different query individuals with distinct term weights search different areas within a document space retrieving partly or totally different set of relevant documents.
The promising outcomes of genetic optimal query search merit continuing research. The areas for future studies also are presented, such as adjusting window size for document retrieval, parallel machine implementation of genetic query search and feature detection of document collection for tuning the outcomes of systems.
| |
MPACT Scores for Jing-Jye Yang A = 0
C = 0
A+C = 0
T = 0
G = 0
W = 0
TD = 0
TA = 0
calculated 2008-01-31 06:23:50
Advisors and Advisees Graph
generating graph, please reload |