多特征方计算优化策略

Optimizing strategies for computing multi-feature cubes

多特征方用于计算复杂的数据挖掘查询,在2n个粒度进行多个依赖的复杂聚集计算。现有的立方体粒度计算技术可以有效计算分布和代数多特征方,针对整体多特征方提出了优化策略:先将立方体水平分块,然后采用冰山查询技术动态选择数据以及局部分布聚集特性优化计算过程。该优化策略既减少了计算复杂度又节省了聚集计算时间,实验结果表明该计算策略比基本的解决方法性能提高一倍以上。

A multi-feature cubes query （MF-Cubes） is a complex data mining query based on data cube, which aggregates on all of the 2^n dependent granularities. There are some efficient algorithms proposed to solve distributive and algebraic MF-Cubes queries. Several strategies for holistic MF-Cubes queries were proposed： 1） horizontal partition of data cube into smaller sub-cubes; 2） a new dynamic subset data selection strategy-Iceberg query; 3） classifying the aggregating functions and identifying the distributive ones, which could use the coarse-finer granularities computing property. These new strategies can simplify the computing complexity and optimize the process, Experiment results illustrate that it has the performance one time higher than the straightforward method.