RoK: Roll-Up with the K-Means Clustering Method for Recommending OLAP Queries

Dimension hierarchies represent a substantial part of the data warehouse model. Indeed they allow decision makers to examine data at different levels of detail with On-Line Analytical Processing (OLAP) operators such as drill-down and roll-up. The granularity levels which compose a dimension hierarchy are usually fixed during the design step of the data warehouse, according to the identified analysis needs of the users. However, in practice, the needs of users may evolve and grow in time. Hence, to take into account the users’ analysis evolution into the data warehouse, we propose to integrate personalization techniques within the OLAP process. We propose two kinds of OLAP personalization in the data warehouse: (1) adaptation and (2) recommendation.

Adaptation allows users to express their own needs in terms of aggregation rules defined from a child level (existing level) to a parent level (new level). The system will adapt itself by including the new hierarchy level into the data warehouse schema. For recommending new OLAP queries, we provide a new OLAP operator based on the K-means method. Users are asked to choose K-means parameters following their preferences about the obtained clusters which may form a new granularity level in the considered dimension hierarchy. We use the K-means clustering method in order to highlight aggregates semantically richer than those provided by classical OLAP operators. In both adaptation and recommendation techniques, the new data warehouse schema allows new and more elaborated OLAP queries.

Our approach for OLAP personalization is implemented within Oracle 10 g as a prototype which allows the creation of new granularity levels in dimension hierachies of the data warehouse. Moreover, we carried out some experiments which validate the relevance of our approach.