Compute Seed Similarities — compute_seed

Compute similarity between response embeddings and SADCAT seed vectors. For each response, computes the correlation (or cosine similarity) between its embedding vector and each SADCAT seed vector. Works for any embedding prefix (SBERT, Gemini, etc.) as long as matching columns exist in both the data and seed vectors.

Usage

compute_seed_similarities(
  data,
  embedding_prefix = "SBERT",
  seed_vectors = Seed_Vectors_Avg,
  method = "correlation",
  response_col = NULL,
  verbose = TRUE
)

Arguments

data: A data.frame with embedding columns (e.g., SBERT_1:SBERT_768)
embedding_prefix: Prefix (or vector of prefixes) identifying embedding columns (default "SBERT"), e.g., c("SBERT", "Gemini").
seed_vectors: Seed vector data. Default: Seed_Vectors_Avg
method: "correlation" (default, uses cor()) or "cosine" (uses lsa::cosine())
response_col: Column used for NA-gating (default: same as text_col, where text_col is inferred as the embedding source column). Only needed if your NA-indicator column differs from the embedding source.
verbose: Print progress? (default TRUE)

Value

The input data with new columns: prefix_SeedName.seed