Group SAX: Extending the Notion of Contrast Sets to Time Series and Multimedia Data

摘要

In this work, we take the traditional notation of contrast sets and extend them to other data types, in particular time series and by extension, images. In the traditional sense, contrast-set mining identifies attributes, values and instances that differ significantly across groups, and helps user understand the differences between groups of data. We reformulate the notion of contrast-sets for time series data, and define it to be the key pattern(s) that are maximally different from the other set of data. We propose a fast and exact algorithm to find the contrast sets, and demonstrate its utility in several diverse domains, ranging from industrial to anthropology. We show that our algorithm achieves 3 orders of magnitude speedup from the brute-force algorithm, while producing exact solutions.