Approximate Similarity Search over Multiple Stream Time Series

Abstract

Similarity search over stream time series has a wide spectrum of applications. Most previous work in static time-series databases and stream time series aim at retrieving the exact answer to a similarity search. However, little work considers the approximate similarity search in stream time series. In this paper, we propose a weighted locality-sensitive hashing (WLSH) technique, which is adaptive to characteristics of stream data, to answer approximate similarity search over stream time series. Due to the unique requirement of stream processing, we present an efficient method to update hash functions adaptive to stream data and maintain hash files incrementally at a low cost. Extensive experiments demonstrate the effectiveness of WLSH, as well as the efficiency of approximate similarity search via hashing on stream time series.