A data mining methodology and its application to semi-automatic knowledge acquisition

We introduce a methodology for knowledge discovery in databases (KDD) where one first discovers large collections of patterns at once, and then performs interactively retrieves subsets of the collection of patterns. The proposed methodology suits such KDD formalisms as association and episode rules, where large collections of potentially interesting rules can be found efficiently. We present methods that support interactive exploration of large collections of rules. With these methods the user can flexibly specify the focus of interest, and also iteratively refine it. We have implemented our methodology in the TASA system which discovers patterns in telecommunication alarm databases. We give concrete examples of how to use frequent patterns in the construction of alarm correlation expert systems.