Efficient Constraint-Based Exploratory Mining on Large Data Cubes

Analysts often explore data cubes to identify anomalous regions that may represent problem areas or new opportunities. Discovery-driven exploration (proposed by S.Sarawagi et al. [5]) automatically detects and marks the exceptions for the user and reduces the reliance on manual discovery. However, when the data is large, it is hard to materialize the whole cube due to the limitation of both space and time. So, exploratory mining on complete cube cells needs to construct the data cube dynamically. That will take a very long time. In this paper, we investigate the optimization methods by pushing several constraints into the mining process. By enforcing several user-defined constraints, we first restrict the multidimensional space to a small constrained-cube and then mine exceptions on it. Two efficient constrained-cube construction algorithms, the NAÏVE algorithm and the AGOA algorithm, were proposed. Experimental results indicate that this kind of constraint-based exploratory mining method is efficient and scalable.