Abstract
In this paper, we focus on the problem of exploring sequential data to discover time sub-intervals that satisfy certain pairwise correlation constraints. Differently than most existing works, we use the deviation from targeted pairwise correlation constraints as an objective to minimize in our problem. Moreover, we include users preferences as an objective in the form of maximizing similarity to users' initial subintervals. The combination of these two objectives are prevalent in applications where users explore time series data to locate time sub-intervals in which targeted patterns exist. Discovering these sub-intervals among time series data is extremely useful in various application areas such as network and environment monitoring.
Towards finding the optimal sub-interval (i.e., optimal query) satisfying these objectives, we propose applying query refinement techniques to enable efficient processing of candidate queries. Specifically, we propose QFind, an efficient algorithm which refines a user's initial query to discover the optimal query by applying novel pruning techniques. QFind applies two-level pruning techniques to safely skip processing unqualified candidate queries, and early abandon the computations of correlation for some pairs based on a monotonic property. We experimentally validate the efficiency of our proposed algorithm against state-of-the-art algorithm under different settings using real and synthetic data.