Exploring multidimensional data cubes using skewness based navigation rules

Navigating through multidimensional data cubes is a non-trivial task. Although On-Line Analytical Processing (OLAP) provides the capability to view multidimensional data through roll-up, drill-down, and slicing-dicing, it offers minimal guidance to end users in the data cube exploration. Currently, a user must have a fairly good understanding of the multidimensional data and a good intuition of what might be discovered in order to navigate through the vast magnitude of combinatorially explosive datasets, which typically include many dimensions, each with multiple levels of dimensional hierarchies. In addition, aggregated views of data often provide incorrect roadmaps for cube navigation since aggregation may hide many interesting low-level details. Also because of large volumes of data in data cubes, it is overwhelming for users to examine the data in tabular representations even for the smaller cuboids. These abovementioned issues create a bottleneck in the data cube exploration.;In this dissertation, we address these data cube exploration problems by proposing DIscovery of Sk-NAvigation Rules (DISNAR), a skewness based algorithm, which discovers hidden patterns in the form of sk-navigation rules using a test of skewness on pairs of current and its candidate 'drill-down' lattice nodes. The discovered rules are then used to provide a method for cube navigation to gain insights into the multidimensional data.;Since a large number of rules may be discovered, making it difficult for users to navigate through the rules, we further introduce three different but complementary measures of interestingness. We first examine the rules for their expectedness of skewness from the neighboring rules. We then introduce an Axis Shift Theory (AST) to identify interesting navigation paths based on the global measures of axis shifts. Lastly, we present an attribute influence approach to identify interesting dimensional attributes providing apriori knowledge about the low-level surprises.;Detailed experimental results show the effectiveness of both the DISNAR algorithm and the measures of interestingness. We conclude that a pruned set of interesting sk-navigation rules offers a much needed and extremely useful guidance to users for exploring multidimensional data cubes, and thus is an interesting research direction.