Text Classification in a Hierarchical Mixture Model for Small Training Sets