Skip to content

v1.4.0 — POEM-based Pareto matching

Choose a tag to compare

@aebrer aebrer released this 06 Mar 03:30
· 24 commits to master since this release
ad0c094

Replace simple Levenshtein matching with POEM (Pareto-Optimal Embedded Matching), a multi-measure consensus algorithm adapted from Brereton et al. 2020.

What's new

  • 6 string similarity measures: Levenshtein, word-sorted Levenshtein, bigram Dice, word Jaccard, substring containment, word containment
  • Pareto front selection with mean distance tiebreak — no single measure can outvote the consensus
  • Top 50% pruning per measure before dominance computation
  • Zero external dependencies
  • "gray sky" now correctly matches Sky Gray instead of Gray Sand

Technical

  • Updated about page documenting the algorithm with published citation
  • PR review fixes: Pareto correctness, consistent Set usage in Jaccard, empty pool guard