Modifying token weights using IDF

[LiquidGunay]

When calculating maxSim we could modify the similiarity score by using the Inverse Document Frequency of the word which the token is a part of, giving it a BM25 esque flavour and try to combine the strenghts of bi-encoders and keyword search. Has anyone tried this approach? I was thinking of this because I haven't found an approach that reliably works on data which is very wordy but also has many numbers.