When the DF of a word is 0.1 It’s like saying that the probability of appearance is 1 in 10 documents. The probability of appearance p is naturally higher the longer the document is. I think it is strange to think that p is a function of the word w only. Wouldn’t it be better to build a model that estimates p from document length n and word w?
This page is auto-translated from /nishio/長さをパラメータにしたDF using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.