Poisson Inverse Regression for sufficient dimension reduction in text data

In this work we present a sufficient dimension reduction (SDR) method for the analysis of text data. We use Poisson distribution to model word occurrences and therefore the method is called Poisson Inverse Regression. We provide the motivation for our example and adjust previously developed algorithms for sufficient dimension reduction for predictors from an exponential family distributions to develop our method. We demonstrate the validity of the proposed method using numerical experiments.