Materialized View Selection for a Data Warehouse Using Frequent Itemset Mining

Data warehouses are subject oriented, consolidated, integrated, and time variant repository of possibly heterogeneous data. A data warehouse is used to response to on-line analytical queries over the millions records of data in an acceptable time. Since a data warehouse often has millions of records of data, it is an important challenge how we can reduce the time of on-line analytical processing. One of the most important issues which address this problem is the view materialization. Each sub-query results an intermediate table, called virtual view, which is used to find final result of the analytical query. These virtual views often are commonly used to response to several analytical queries. We can materialize such views to prevent multiple redundant computations and thus lead to reduction in response time of queries. The constraint of storage memory on one hand, and the maintenance cost of materialized views when the source data are updated on the other hand, cause that it is impossible to materialize all or even large part of views. Therefore, selection of a proper set of views to materialization plays a major role in performance. There are many methods of view selection to materialization which uses different techniques and frameworks to select optimal set of views to materialization. In this paper, we present a new efficient method to conduct selecting proper set of views to materialization using a frequent itemset mining approach. In our algorithm, the set of given queries is transformed to a transaction database where a transaction corresponds to a query and items of a transaction are the original query's predicates. Our performance study showed that this algorithm outperformed substantially the best former algorithms.