in any training document pertaining to a particular category c will otherwise render P (c|d) zero. 3 Unbalanced Class Sizes Many text categorization problems are unbalanced. This can cause problems because [...] classes after normalization. To do this we replace nwd, where d is in Dc, by α × nwd∑ w′ ∑ d∈Dc nw′d , (3) i.e. we normalize the vector of word counts for each class to have length α when measured according [...] classifier for text classification. 4 Relationship to Centroid Classifier In the centroid classifier [3], each class is represented by its mean word vector, normalized to unit length using L2 norm. The centroid …