Reproduce “Matchmaker: Self-Improving Large Language Model Programs for Schema Matching” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Nabeel Seedat, Mihaela van der Schaar

A self-improving, compositional large language model program, Matchmaker, automates schema matching, outperforming existing methods on real-world medical benchmarks. The approach facilitates data integration by operating effectively in zero-shot settings and enhancing the creation of ML-ready datasets.

Key insights

  • Decomposing schema matching into a multi-stage compositional LLM program allows for leveraging diverse reasoning capabilities and improves performance.
  • Zero-shot self-improvement through systematically generated synthetic in-context examples can effectively optimize complex LLM programs without human-labeled data.
  • Integrating both semantic retrieval and LLM-based reasoning for candidate generation captures complex schema relationships more effectively than relying on semantic similarity alone.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.