Layered relevance scoring for portal connectors (title + long text + CPV)

Scores a tender tibble the way the VMP-BB pipeline does, but in one call for connectors that already ship a description and CPV codes (e.g. the API portals): title_cols use the full rule (>=1 strong OR >=2 supporting), text_cols (long free text) are matched STRONG-only (incidental supporting hits in long text are noise), and cpv_col codes are mapped to groups. The three group sets are merged into groups, with match_source (title/detail/cpv), cpv_groups, score and is_relevant.

Usage

score_layered(
  df,
  title_cols,
  text_cols = character(),
  cpv_col = NULL,
  keywords = tender_keywords(),
  cpv_map = tender_cpv_map(),
  exclude = TRUE
)

Arguments

df

A data frame of tenders.

title_cols

Columns scored with the full rule (e.g. the title).

text_cols

Columns scored strong-only (e.g. description); default none.

cpv_col

Name of a comma/space-separated CPV column, or NULL.

keywords

Keyword groups (default tender_keywords()).

cpv_map

CPV-to-group map (default tender_cpv_map()).

exclude

Apply apply_title_excludes() afterwards to drop construction / building / maintenance tenders (default TRUE).

Value

df with groups, cpv_groups, match_source, score, is_relevant added, sorted by descending score.