From matched pairs to one row per item: grouping matches
A list of matched pairs is not a clean item list. How to turn pairs into groups with an id, a size and a shape, and why to build them after review.
A list of matched pairs is not a clean item list. How to turn pairs into groups with an id, a size and a shape, and why to build them after review.
Comparing a list with itself grows with the square of its size. A recipe that finds near-duplicates without listing every pair, then removes them.
An AI check has a language model read two records and answer your questions about them. How it works, how to route doubtful pairs to people, when to trust it.
Picking each record's top candidate lets two records claim the same partner. Four ways to enforce one-to-one links, measured on 1,000 × 1,150 company names.
Matching resumes to job descriptions takes more than keyword overlap. Build a multi-signal workflow that handles thousands of candidates and hundreds of roles.
Talent matching consultants spend more time building pipelines than matching candidates. Configurable matching infrastructure changes the math.
A cited timeline of record matching, from Russell's 1918 Soundex patent through Fellegi-Sunter and Jaro-Winkler to embeddings and LLMs, and what each left open.
SQL JOINs and pandas merges fail on color variants, promo naming, translations and spec formats. See where traditional tools hit a ceiling and what works.
How to handle the N x M explosion in record matching — blocking strategies, pre-filter cascades, batch processing, and fault tolerance for large datasets.
Matching slows at scale. Learn how blocking keys cut comparisons by orders of magnitude, how to choose them, and how multi-pass blocking recovers misses.
From pandas merge to fuzzywuzzy to recordlinkage: a practical guide to CSV matching in Python, and how to tell when custom code stops making sense.
How embedding-based record matching works: cosine similarity, threshold tuning, pair counts, what to embed, failure modes, and how to test it on your data.
A six-step procedure for picking a record matching method: tag a labelled sample by difference type, map types to methods, block, and tune on a hold-out.
Addresses are the hardest field to match: abbreviations, unit numbers, directionals and international formats defeat exact matching. Here's how to handle them.
A practical breakdown of six fuzzy matching algorithms — how they work, where they excel, and when to combine them for record matching across messy datasets.
Deterministic vs probabilistic record matching, defined and tested on 1.15M synthetic company-name pairs: what each method finds, misses, and when to use it.