Maximizing the Early Abandon Effect in Time-Series Similar Sequence Matching

초록

In recent years, there have been many efforts on exploiting a large time-series database, and their major research topic is similar sequence matching that identifies data sequences similar to a query sequence. In this paper, we address the problem of maximizing the early abandon effect in computing the Euclidean distances for similar sequence matching. The early abandon improves the matching performance by stopping the computation process immediately after the intermediate distance exceeds a user-specified tolerance. We observe that the starting offset highly influences the early abandon effect, and we thus try to select the starting offset so as to maximize the early abandon effect. We first propose MaxOffset that uses the maximum entry of a query sequence as its starting offset. As an extension of MaxOffset, we then propose BiDirection that considers both directions of the maximum entry, i.e., left-side adjacent entries as well as right-side adjacent entries. The intuition behind these algorithms is that a large portion of the actual distance might be accumulated around maximum entries. We empirically showcase the superiority of the proposed algorithms.

키워드