Research on domain-driven actionable knowledge discovery

Abstract

Traditional data mining is a data-driven trial-and-error process, stop on general pattern discovered. However, in many cases the mined knowledge by this process could not meet the real-world business needs. Actually, in real-world business, knowledge must be actionable, that is to say, one can do something on it to profit. Actionable knowledge discovery is a complex task, due to it is strongly depend on domain knowledge, such as background knowledge expert experience, user interesting, environment context, business logic, even including law, regulation, habit, culture etc. The main challenge is moving data-driven into domain-driven data mining (DDDM), its goal is to discover actionable knowledge rather than general pattern. As a new generation data mining approach, main ideas of the DDDM are introduced. Two types of process models show the difference between loosely coupled and tightly coupled. Also the main characteristics, such as constraint-base, human-machine cooperated, loop-closed iterative refinement and meta-synthesis-base process management are proposed. System architecture will be introduced, as well as a paradigm will be introduced. © 2009 Springer Berlin Heidelberg.