Advances in Spatial and Temporal Databases

Volume 6849 of the series Lecture Notes in Computer Science pp 422-440

Quality of Similarity Rankings in Time Series

  • Thomas BerneckerAffiliated withLudwig-Maximilians-Universität München
  • , Michael E. HouleAffiliated withNational Institute of Informatics
  • , Hans-Peter KriegelAffiliated withLudwig-Maximilians-Universität München
  • , Peer KrögerAffiliated withLudwig-Maximilians-Universität München
  • , Matthias RenzAffiliated withLudwig-Maximilians-Universität München
  • , Erich SchubertAffiliated withLudwig-Maximilians-Universität München
  • , Arthur ZimekAffiliated withLudwig-Maximilians-Universität München

* Final gross prices may vary according to local VAT.

Get Access


Time series data objects can be interpreted as high- dimensional vectors, which allows the application of many traditional distance measures as well as more specialized measures. However, many distance functions are known to suffer from poor contrast in high-dimensional settings, putting their usefulness as similarity measures into question. On the other hand, shared-nearest-neighbor distances based on the ranking of data objects induced by some primary distance measure have been known to lead to improved performance in high-dimensional settings. In this paper, we study the performance of shared-neighbor similarity measures in the context of similarity search for time series data objects. Our findings are that the use of shared-neighbor similarity measures generally results in more stable performances than that of their associated primary distance measures.