Skip to contents

Apply semantic suggestions into a dictionary

Source: R/semantics-helpers.R
apply_semantic_suggestions.Rd

Copies selected IRIs from a semantic_suggestions tibble into the matching dictionary fields. Suggestions remain separate by default; this helper gives you an explicit merge step when you decide the top candidates are good enough.

Usage

apply_semantic_suggestions(
  dict,
  suggestions = attr(dict, "semantic_suggestions"),
  strategy = c("top", "reviewed", "llm"),
  columns = NULL,
  roles = NULL,
  min_score = NULL,
  min_llm_confidence = NULL,
  overwrite = FALSE,
  verbose = TRUE
)

Arguments

dict

A dictionary tibble, typically returned by infer_dictionary() or suggest_semantics().

suggestions

A suggestions tibble, usually attr(dict, "semantic_suggestions"). If omitted, the function reads that attribute from dict.

strategy

Selection strategy per column-role pair. "top" keeps the original lexical ranking. "reviewed" applies only rows whose decision is "accepted" (or the equivalent "accept"). "llm" applies only candidates marked with llm_selected = TRUE by suggest_semantics(..., llm_assess = TRUE). When reviewed or LLM-reviewed selections contain multiple constraints for one measurement, their IRIs are deduplicated in first-occurrence order and written to constraint_iri as the SDP-compatible semicolon-separated value. Other roles continue to select one value per column-role pair.

columns

Optional character vector limiting application to specific column_name values.

roles

Optional character vector limiting application to specific suggestion roles: "variable", "property", "entity", "unit", "constraint", "statistical_modifier".

min_score

Optional numeric threshold. Only available when suggestions includes a score column; otherwise the function errors.

min_llm_confidence

Optional numeric threshold for strategy = "llm". Requires llm_confidence in suggestions.

overwrite

Logical; if FALSE (default), only missing fields are filled. Set TRUE to intentionally replace existing IRIs.

verbose

Logical; if TRUE (default), print a short summary.

Value

The dictionary tibble with selected semantic IRI fields filled in.

Details

Matching is done by both column_name and dictionary_role. When the suggestions tibble also includes dataset_id and table_id, those keys are honored too. Suggestions that target non-column destinations (for example codes.csv, tables.csv, or dataset.csv) are ignored by this helper and remain review-only.

Examples

dict <- infer_dictionary(data.frame(spawner_count = c(12L, 19L)))
# Example-only IRI: replace it with a reviewed term for a real SDP.
reviewed <- tibble::tibble(
  column_name = "spawner_count", dictionary_role = "variable",
  iri = "https://example.org/terms/spawner-count", score = 3
)
filled <- apply_semantic_suggestions(
  dict, suggestions = reviewed, columns = "spawner_count",
  min_score = 2, verbose = FALSE
)
filled$term_iri
#> [1] "https://example.org/terms/spawner-count"