Text vectorizer that capture feature offset in the text?

I’m using sklearn Tfifdfvectorizer to extract feature from text towards text classification.
I believe the information I need tends to be in the beginning of the document, so I would like to somehow capture the offset of each feature per document (either of the first appearance, or the mean offset over of all appearances).
Is there some vectorizer that can do that? or some other method of extracting this information efficiently?

Thank you!