Back
to indexFrancisco Pereira and Ernesto Costa
Summary
This article proposes and analyzes an adaptive agent approach to online searching. However, its algorithms and approaches do not diverge from previous work from Menczer and Belew on InfoSpiders. Its addition to the existing body of knowledge includes a rather useless learning percentage, which is not coorelated with the learning parameter. Apparently, the learning percentage is a probability that the learned data will be used. This seems to have little impact on the overall performance of the agent. Another interesting addition to the basic evolutionary algorithm is the use of the weighted sum of syntactic analysis, syntactic quality, and document quality to determine which link should be followed.
Methods
One interesting method includes the determination of the link potential, which is the independantly weighted sum of the syntactic analysis (word neighborhood), syntactic quality (similarity to relevant subject) and document quality. This is used in the place of a neural net for the link determination. Experimental evidence shows that this approach works only partially.
Keywords
Artificial life, evolution, learning, syntactic analysis, syntactic quality, document quality, learning percentage
Rating
3
Bibtex Entry
@misc{ ernesto,
author = "Francisco Pereira and Ernesto Costa",
title = "How Learning Improves the Performance of Evolutionary Agents: a Case Study with an Information Retrieval System for a Distributed Environment",
url = "citeseer.nj.nec.com/421484.html"
}