The Online Mining Of Data Cube Gradient

Cube gradients are generalization of association rules which represent how a set of measures is affected by modifying a cube cell's dimensions. Cube gradients are signifcantly more expressive than association rules in capturing trends and patterns in data, which can support sophisticated "what if" analysis tasks.In order to overcome the greate compute cost cased by the dominating cube gradient mining algorithms which usually compute all possible cube gradients from srouce table directly, the algorithm of mining cube gradient online, MCGO is proposed. Because all cube gradients are containted in data cube that has been materialized in online analytical processing servers to accelerate the response, MCGO first acquires cube cells by queries to data cube, and then computes cube gradients from the cube cells by the dimension value matching and gradient constraints judging. It not only saves the greate compute cost, but also get rid of the redundancies between cube gradients and cube cells. That the mining of cube gradient is integrated into online analytical processing also accords with users' interest arising when browsing the data cube.The huge size of data cube may make the storage and maintenance of data cube become unpractical, while condensed cube can save storage space greatly, based on which the cube index CuboidTree can improve query efficiency. The algorithm of mining cube gradient online based on condensed cube, MCGOBC is proposed. Noticed the fact that those cube tuples aggregated from the same base tuple set are condensed into one physical tuple and the base single tuples with same single dimension set are organized into a virtual cuboid in CuboidTree, MCGOBC prunes virtual cuboid queries and expands base single tuples directional. Those measures assure that the virtual cuboid only be queried once and base single tuples only be expanded out all possible needed cube cells, which helps to save query spending and improve the algorithm performance. After the online analytical mining model is studied, an online cube gradient mining tool DM_MCGOBC is designed based on the online analytical processing server DM_OLAP.