- author = "Ronan Cummins",
- title = "The Evolution and Analysis of Term-Weighting Schemes in Information Retrieval",
- school = "National University of Ireland, Galway",
- year = "2008",
- address = "Ireland",
- month = may,
- keywords = "genetic algorithms, genetic programming",
-
URL = "
http://www3.it.nuigalway.ie/cirg/rcummins_thesis.pdf",
- size = "201 pages",
-
abstract = "Information Retrieval is concerned with the return of
relevant documents from a document collection given a
user query. Term-weighting schemes assign weights to
keywords (terms) based on how useful they are likely to
be in identifying the topic of a document and are one
of the most crucial aspects in relation to the
performance of Information Retrieval systems. Much
research has focused on developing both term-weighting
schemes and theories to support them.
Genetic Programming is a biologically-inspired search algorithm useful for searching large complex search spaces. It uses a Darwinian-inspired survival of the fittest approach to search for solutions of a suitable fitness. This thesis outlines experiments that use Genetic Programming to search for term-weighting schemes. A study of term-weighting schemes in the literature is undertaken and consequently, the function space is separated into three areas that represent three fundamental concepts in term weighting.
Experiments using Genetic Programming to search these three function spaces show that term-weighting schemes that outperform state of the art term-weighting benchmarks can be found. These experiments also show that the new term-weighting schemes have general properties as they achieve high performance on unseen test data. An analysis of the solution space of the term-weighting schemes shows that the evolved solutions exist in a different part of the space than the current benchmarks. These experiments show that the Genetic Programming approach consistently evolves solutions that return similar ranked lists in each of the three function spaces.
Furthermore, the best performing term-weighting schemes are formally analysed and are shown to satisfy a number of axioms in Information Retrieval. A detailed analysis of the existing axioms is presented together with some amendments and additions to the existing axioms. This analysis aids in theoretically validating the term-weighting schemes evolved in the framework.
Finally, a secondary application of Genetic Programming to Information Retrieval is presented to show the potential for Genetic Programming in addressing other issues in Information Retrieval. This experiment shows that Genetic Programming can be used to combine further evidence in the retrieval process to enhance performance. This approach evolves schemes for use with two automatic query expansion techniques to increase retrieval effectiveness.",
- notes = "Supervisor: Colm O'Riordan",
