摘要：We introduce a new method for building classi cation models when we
have prior knowledge of how the classes can be arranged in a hierarchy, based on
how easily they can be distinguished. The new method uses a Bayesian form of
the multinomial logit (MNL, a.k.a. \softmax") model, with a prior that introduces
correlations between the parameters for classes that are nearby in the tree. We
compare the performance on simulated data of the new method, the ordinary MNL
model, and a model that uses the hierarchy in a di
erent way. We also test the
new method on page layout analysis and document classi cation problems, and
nd that it performs better than the other methods.
