Efficient Computation Of Iceberg Cubes With Complex Measures

Abstract

It is often too expensive to compute and materialize a complete high-dimensional data cube. Computing an iceberg cube, which contains only aggregates above certain thresholds, is an effective way to derive nontrivial multi-dimensional aggregations for OLAP and data mining. In this paper, we study efficient methods for computing iceberg cubes with some popularly used complex measures, such as <i>average</i>, and develop a methodology that adopts a weaker but anti-monotonic condition for testing and pruning search space. In particular, for efficient computation of iceberg cubes with the <i>average</i> measure, we propose a <i>top-k average</i> pruning method and extend two previously studied methods, Apriori and BUC, to Top-<i>k</i> Apriori and Top-<i>k</i> BUC. To further improve the performance, an interesting hypertree structure, called H-tree, is designed and a new iceberg cubing method, called Top-<i>k</i> H-Cubing, is developed. Our performance study shows that Top-<i>k</i> BUC and Top-<i>k</i> H-Cubing are two promising candidates for scalable computation, and Top-<i>k</i> H-Cubing has better performance in most cases.