ABSTRACT
In this chapter, we discuss different mathematical formulations of the NMF model that vary according to the objective function minimized in the underlying constrained optimization problem. The effects of these formulations on the computability and quality of the resulting matrix factors needed for document interpretation (i.e., the
metadata) and clustering are discussed. The success of future document classification systems based on such factor analytic approaches will greatly depend on how well the corresponding mathematical formulations capture both obvious and latent semantic information subject to time and space constraints of the available computing environment.
