Population surname-rarity weight
population_surname_weight.RdThe cross-organization analogue of the within-organization surname_weight().
Within an organization the question is "how many distinct people here share
this surname" (a family board); across organizations the question is "how rare
is this surname in the population" – two people named SMITH agreeing across
orgs is weak evidence, two named GANTSOUDES is strong. Returns a normalized
inverse-document-frequency weight per surname: log(N / n_s) / log(N), where
N is the number of people and n_s the number whose canonical surname is
s. A unique surname scores ~1; a surname everyone shares scores ~0.
Arguments
- profiles
Output of
build_person_profile()(uses the canonicallast_name, one per person).