When acquiring knowledge from data, it has been found that, in complex domains, many individual data mining techniques fail to find adequate concept descriptions for all of the concepts contained in the data. Rather, the techniques usually find complementary results with respect to these concepts. If two or more techniques can cooperate when constructing a knowledge repository, the methods may teach one another to recognize additional concepts. In this way, the quality of the final knowledge repository should be enhanced.

This talk concerns a framework consisting of data mining agents that cooperate in a multi-agent learning system. The cooperative inductive learning team (CILT) approach is presented, followed by a discussion of the lessons learned when applying this approach to a number of real-world data repositories.