Use for sequence classification and labeling
Hidden Markov models for sequence labeling

The task of sequence classification is often distinguished from the task of sequence segmentation (or labeling). To design a sequence classification system we have a learning base B = {(x^k, y^k), k = 1... K} where y^k ∊ {U₁, U₂,..., U_M} denotes the sequence class.

A sequence classification system based on hidden Markov models is learned by maximizing a criterion. The simplest and most widely used is the maximum likelihood criterion, but other criteria can also be used (see § 5.2