We propose methods for incrementally constructing a knowledge model for a dynamically changing database, using a swarm of special agents (ie an ant colony) and imitating their natural cluster-forming behavior. We use information-theoretic metrics to overcome some inherent problems of ant-based clustering, obtaining faster and more accurate results. Entropy governs the pick-up and drop behaviors, while movement is guided by pheromones. The primary benefits are fast clustering, and a reduced parameter set. We compared the method both with static clustering (repeatedly applied), and with the previous dynamic approaches of other authors. It generated clusters of similar quality to the static method, at significantly reduced computational cost, so that it can be used in dynamic situations where the static method is infeasible. It gave better results than previous dynamic approaches, with a much-reduced tuning parameter set. It is simple to use, and applicable to continuously- and batch-updated databases.
[1]
Ajith Abraham,et al.
Swarms on continuous data
,
2003,
The 2003 Congress on Evolutionary Computation, 2003. CEC '03..
[2]
Juan Julián Merelo Guervós,et al.
Self-Organized Stigmergic Document Maps: Environment as a Mechanism for Context Learning
,
2004,
ArXiv.
[3]
Jean-Louis Deneubourg,et al.
The dynamics of collective sorting robot-like ants and ant-like robots
,
1991
.
[4]
Yi Li,et al.
COOLCAT: an entropy-based algorithm for categorical clustering
,
2002,
CIKM '02.
[5]
Baldo Faieta,et al.
Diversity and adaptation in populations of clustering ants
,
1994
.
[6]
Marco Dorigo,et al.
Swarm intelligence: from natural to artificial systems
,
1999
.