Distributional clustering of words for text categorization