Is it Enough to Simply Apply Language Model for Optimal Text Classification?