Model-Based Similarity Measure in TimeCloud

This paper presents a new approach to measuring similarity over massive time-series data. Our approach is built on two principles: one is to parallelize the large amount computation using a scalable cloud serving system, called TimeCloud. The another is to benet from the lter-and-renement approach for query processing, such that similarity computation is eciently performed over approximated data at the lter step, and then the following renement step measures precise similarities for only a small number of candidates resulted from the ltering. To this end, we establish a set of rm theoretical backgrounds, as well as techniques for processing kNN queries. Our experimental results suggest that the approach proposed is ecient and scalable.