Toward measuring the similarity of complex event sequences in real-time

Abstract

Traditional sequence similarity measures have a high time complexity and are therefore not suitable for real-time systems. In this paper, we analyze and discuss properties of sequences as a step toward developing more efficient similarity measures that can approximate the similarity of traditional sequence similarity measures. To explore our findings, we propose a method for encoding sequence information as a vector in order to exploit the advantageous performance of vector similarity measures. This method is based on the assumption that events closer to a point of interest, like the current time, are more important than those further away. Four experiments are performed on both synthetic and real-time data that show both disadvantages and advantages of the method. © 2012 Springer-Verlag.