Great article. Curious: Why is polynomial kernel particularly popular in NLP (whereas RBF kernel seems to be more of the go-to choice for SVM classification tasks outside NLP--assuming manageable data sizes?)