Abstract
In a human learning, humans do not always use one kind of knowledge representation. They use a concrete knowledge representation at first, but they gradually come to use a abstract one as they learn more. We have studied a system that reasons with a knowledge representation and switches to another one when some conditions are satisfied.
Humans reason with abstract knowledge representation for the data that have a lot of similar data but concrete one for the data that have a few similar data. They choose a knowledge representation according to the data. While the system reasons with the current knowledge representation for any data. In this paper, therefore, we propose that the system use knowledge representation of a different level of abstraction according to the numbers of similar data, where it generates clusters for the given data set and associates a knowledge representation with each cluster. The result is compared with that for the current system.
