Abstract
Data analysts arduously rely on data visualizations for drawing insights into huge and complex datasets. However, finding interesting visualizations by manually specifying various parameters such as type, attributes, granularity is a protracted process. Simplification of this process requires systems that can automatically recommend interesting visualizations. Such systems primarily work first by evaluating the utility of all possible visualizations and then recommending the top-k visualizations to the user. However, this process is achieved at the hands of high data processing cost. That cost is further aggravated by the presence of numerical dimensional attributes, as it requires binned aggregations. Therefore, there is a need of recommendation systems that can facilitate data exploration tasks with the increased efficiency, without compromising the quality of recommendations. The most expensive operation while computing the utility of the views is the time spent in executing the query related to the views. To reduce the cost of this particular operation, we propose a novel technique mView, which instead of answering each query related to a view from scratch, reuses results of the already executed queries. This is done by incremental materialization of a set of views in optimal order and answering the queries from the materialized views instead of the base table. The experimental evaluation shows that the mView technique can reduce the cost at least by 25-30% as compared to the previously proposed methods.