Handling KDD process iteration by integration of planning and controlling

In our GLS (global learning scheme) system, KDD (knowledge discovery in databases) process is modeled as an organized society of intelligent agents (called KDD agents). KDD process planning is a meta-agent in this framework, and some basic AI planning techniques have been used to organize dynamically the KDD process so that the GLS system increases both autonomy and versatility. In this paper, we address some deeper issues concerning KDD process planning. Especially we propose a mechanism integrating process planning with process controlling to formalize process iteration (that is, the re-execution of some agents). In this mode, execution failures causing process iteration, feedback paths and re-execution of the agents in the paths (and even the iterating number for each loop) are determined dynamically and automatically. With this and other deeper issues (such as process evolution) being properly handled, the GLS system is more complete in KDD process modeling, and more flexible and robust for practical use.