Add acronym and prefix recovery columns to a normalized frame

Adds three blocking columns:

name_acro

name_key with the record's initials appended as extra synthetic tokens. Blocking on this column lets a query that is an acronym (NJOP) or merely contains one (PVCC EDUCATIONAL FOUNDATION) find the spelled-out reference record. Initials are rare, so they carry high IDF and justify a pair on their own — while the IDF machinery still prunes the common ones.

name_pre

the first prefix_len characters of name_key, for the truncation pass.

name_is_acro

TRUE when the whole name is a single short all-letter token, i.e. the query is itself an acronym.

Usage

pf_add_recovery_keys(d, prefix_len = 18L, only_truncated = NULL)

Arguments

d

A normalized frame from npmatch::np_normalize().

prefix_len

Characters to keep for the truncation prefix key.

only_truncated

Optional logical vector, one per row. When supplied, name_pre is NA wherever it is FALSE, so the prefix pass never fires on a name that was not cut off.

Value

d with the three columns added.

Details

Apply to both sides. On the reference it supplies the initials to match against; on the query it supplies the prefix and the acronym flag.

Restricting the prefix key

On the reference side every record gets a name_pre. On the query side it should be set only for names that are actually truncated — pass only_truncated. Without that restriction the pass fires on every long name and an 18-character prefix is far too blunt a key: AMERICAN ASSOCIATION OF PRO-LIFE OBSTETRICIANS and AMERICAN ASSOCIATION OF ZOO KEEPERS share one, and that exact false positive appeared in testing.