SIREN: A Python Package for Automated Psychometric Scale Shortening Using Document Embeddings

SIREN (Semantic Item Reduction Engine) is an open-source Python package that shortens unidimensional and multidimensional scales from item text alone, without respondent data. Items are encoded as transformer-based embeddings, and scale reduction is framed as constrained optimization over the resulting similarity matrix: retained items maximize within-dimension coherence or coverage while minimizing between-dimension similarity, subject to user-specified item counts per dimension. Ant colony optimization is the default search, with SLSQP, genetic algorithm, and simulated annealing alternatives. Diagnostic functions report coverage of dropped items, benchmark any short form against random subsets, and flag redundant items. SIREN runs on Windows, Linux, and macOS and is installable from PyPI.