This notebook illustrates some of the behaviors of the singular value decomposition of time series data. I’ve written it in part as a response to a paper by Andrew J. Reagan, Lewis Mitchell, Dilan Kiley, Christopher M. Danforth, and Peter Sheridan Dodds recapitulating an argument first made by Matt Jockers: that certain kinds of eigendecompositions of sentiment data from works of fiction will reveal “fundamental plot structures” or “fundamental narrative arcs.”
Read full post here.