In traditional bloom filters, each item is hashed some fixed number of locations. One variant of this is to hash items a varying number of locations within the same bloom filter.
This idea is presented in Zhong et al., Optimizing data popularity conscious bloom filters, PODC '08 and Bruck et al., Weighted Bloom filter, IEEE Information theory '06. However, I know of an independent and roughly contemporary "discovery" of this idea in industry that wasn't published.
This idea seems like the kind of thing that may have been independently discovered multiple times and I'd like to find the earliest public reference to cite. It's possible this exists somewhere in the IR literature as a variant of "signature files", or elsewhere with terminology that I'm not familiar with.
In the IR literature, there's Faloutsos et al., Design of a signature file method that accounts for non-uniform occurrence and query frequencies, VLDB '85, which is quite close but if I'm reading the paper correctly the paper only does the math for splitting terms into two separate classes and doesn't generalize the math like Zhong et al. or Bruck et al.