Automated Explanations of User-Expected Trends for Aggregate Queries.

Recently, a deeper level of data exploration has emerged enabling users to infer anomalies in their queries. This exploration level strives to explain why a particular anomaly exists within a query result by providing a set of explanations. These explanations are precisely a set of alterations, such that when applied on the original query cause anomalies to disappear. Trends are pattern changes in business applications generated based on SQL aggregated queries. Additionally, a user expected trend is a particular pattern change in data was supposedly happen based on businesses studies. In this paper, we generalize this process to automatically produce explanations for users expected trends. We propose User Trend Explanations (UTE) framework which provides insightful explanations by taking a set of user-specified points (called prospective trend), and finds a top explanation that produce this trend. We develop a notion of uniformity of a predicate on a given output, and implement a set of algorithms to search the data space efficiently and effectively. The key idea is harnessing the linear search space rather than the exponential space to enable accurate explanations that are possible with tuples. Our experiments on real datasets show significant improvements UTE provides when compared with state-of-the-art related algorithms.