Compute TFIDF iteratively

[minottic]

I think that at the cost of some storage space, the TFIDF score can be computed iteratively, without having to run the whole computation from the beginning when a new document is added.

The idea is:

1. for each (item, term) you store its TF at time T and TFIDF at time T
2. you store Tc at time T
3. for each term you store T(t) at time T
   now let's say that at time T+1 you add one new item
4. you increment Tc by 1 and store it
5. for each term you increment T(t) by 1 if the new document contains t and store it
6. with 6 you compute TF and TFIDF for the new document at time T+1 and store
7. with 6 you update the TF and TFID of the old documents at T+1

All this only makes sense if I understood correctly how TFIDF works :)