Interactive Mining on Hierarchical Data

interactive, intuitive, on-line, data mining, OLAP, data warehouse, association, characterization, classification, nonnaïve Bayessian classification, uml notation based presentation

In this paper, we propose a framework for interactive, iterative, and intuitive mining of multilevel association, characterization and classification rules on data organized in multi-level conceptual hierarchies. This framework is called OLAM SE (Self Explaining On-Line Analytical Mining) and it is proposed as an extension of OLAP or as an alternative to Han's OLAM. OLAM processes data stored in data cubes structure of which is based on a given conceptual hierarchy. OLAM SE determines minimum support value from user defined cover value of data with usage of entropy coding principle. It also automatically determines the maximum threshold to avoid explaining knowledge that is obvious and so potentially uninteresting. Major part of data is thus described by frequent patterns. The presentation of results is inspired by UML diagram notation. It contains a graph nodes of which are frequent data sets represented as packages including sub packages - data classes or items. Edges represent relations or patterns between packages. This representation could be applicable for characterization and nonnaïve Bayesian classification process as well. Patterns can be interactively explored by the user, who gets a detailed view of attractive ones. She can intuitively drive the more detailed knowledge obtaining process.