On Novelty Evaluation of Potentially Useful Patterns

As is generally accepted, the most important feature that a Knowledge Discovery in Database (KDD) system must possess is, to be able to discover patterns that are “novel” and “potentially useful”. In order to allow KDD systems to make novelty and potential usefulness judgment, we extend our former work on discovering “potentially useful” patterns by proposing a formal definition of “novelty” based on the same probabilistic logic foundation we used to define “potential usefulness”. Furthermore, a tractable algorithm is proposed that is capable of discovering all novel and potentially useful patterns from databases based upon limited accessible information.