In this paper we propose a new operator for advanced exploration of large multidimensional databases. The proposed operator can automatically generalize from a specific problem case in detailed data and return the broadest context in which the problem occurs. Such a functionality would be useful to an analyst who after observing a problem case, say a drop in sales for a product in a store, would like to find the exact scope of the problem. With existing tools he would have to manually search around the problem tuple trying to draw a pattern. This process is both tedious and imprecise. Our proposed operator can automate these manual steps and return in a single step a compact and easy-to-interpret summary of all possible maximal generalizations along various roll-up paths around the case. We present a flexible cost-based framework that can generalize various kinds of behaviour (not simply drops) while requiring little additional customization from the user. We design an algorithm that can work efficiently on large multidimensional hierarchical data cubes so as to be usable in an interactive setting.
展开▼
