Mining Interestingness Sub-cubes in Multi-dimensional Data

When dealing with the multi-dimensional data, despite all the data cells presented, users may only be interested in those that satisfy some condition. Now that the desired cells are scattered around, our goal is to identify the sub-cubes with high proportion of desired cells, which shed light on the distributions that users are most interested in, which we denote as interestingness sub-cubes. With all of the cells satisfying the given condition, the cube is called closed interestingness sub-cube (CIS). With a few undesired cells being involved, the cube turns flowed interestingness sub-cube (FIS) which may represent users´ interest more generally and accurately. In this paper, we study the problem of CIS mining and FIS mining. For CIS mining, a 3D frequent closed cube mining algorithm is extended to suit multi-dimensional circumstance. For FIS mining, a bottom-up and top- down combined method is proposed to rapidly locate the possible interestingness sub-cubes followed by heuristic cutting. Our performance study shows that both are effective and efficient.