IDF-weighted containment of a purpose in a document

The share of the purpose's information content that is present in the organization's text: $$ \mathrm{cov}(q,d) = \sum_{t \in q \cap d} \mathrm{idf}(t) \;/\; \sum_{t \in q} \mathrm{idf}(t) $$

Usage

pf_purpose_coverage(purpose, doc, idf)

Arguments

purpose, doc

Character vectors, recycled to common length.

idf

Named numeric of token IDF weights (see pf_build_org_text()).

Value

A numeric vector in [0, 1], with NA where undefined.

Details

Asymmetric on purpose. The query is two words and the document can be five thousand, so a symmetric measure (cosine, Jaccard) is near zero for every pair and cannot rank them; containment asks only how much of the query is accounted for, and is insensitive to document length. NA when the purpose has no content tokens or the document is empty — never 0, because a zero is evidence against and this is an absence of evidence.