The shocklet transform: A decomposition method for the identification of local mechanism-driven dynamics in sociotechnical time series

David Rushing Dewhurst, Thayer Alshaabi, Dilan Kiley, Michael V. Arnold, Joshua R. Minot, Christopher M. Danforth, and Peter Sheridan Dodds



We introduce an unsupervised pattern recognition algorithm termed the Discrete Shocklet Transform (DST) by which local dynamics of time series can be extracted. Time series that are hypothesized to be generated by underlying deterministic mechanisms have significantly different DSTs than do purely random null models. We apply the DST to a sociotechnical data source, usage frequencies for a subset of words on Twitter over a decade, and demonstrate the ability of the DST to filter high-dimensional data and automate the extraction of anomalous behavior.

Online Appendices

Appendix A (see the paper)
Online Appendix A - Ranked shock indicator plots (2009-2018)
Online Appendix B - Ranked spike indictor plots (2009-2018)
Online Appendix C - Individual comparisons for over 10000 1gram time series with STAR and Twitter's ADV

Last updated: