Record Linkage Resources

Resources for tackling record linkage (also known as deduplication, data matching, entity resolution)

Note: If you're looking for file deduplication software, you're in the wrong place! This page focuses on deduplicating datasets.

Also note: Nor is this page is not about deduplication software used in backup and storage.

Record linkage attempts to identify duplicate records in messy data. It is a thorny problem that crops up in a variety of scenarios that attempt to understand with real-world entities (most often people), such as census and statistical bureaus, medical organizations, the social sciences, and of course commercial business.

For example, are these records the same person? Record Linkage is how you make the computer decide--quickly.

Background

Documents

• Wikipedia pages on Record Linkage and Data Deduplication
• Overview slides: https://www.umiacs.umd.edu/~getoor/Tutorials/ER_VLDB2012.pdf
• Dedupe's explanation of how their software works: https://dedupe.io/developers/library/en/latest/How-it-works.html

Talks

• Flávio Juvenal, 1 + 1 = 1 or Record Deduplication with Python: https://www.youtube.com/watch?v=4O87RdBgRJ4 (2018)
• Peter Christen, Record Linkage lectures at ADRC-Scotland: 1 2 3 4 (2015)
• Mike Mull, The Art and Science of Data Matching: https://www.youtube.com/watch?v=Y-nYEOgq3YE (2015)
• Rhydwyn Mcguire, Join for real life: https://www.youtube.com/watch?v=cEcVIjyHfiQ (2013)
• Andrew Rowe, Big Data Deduplication and Data Matching using Python: https://www.youtube.com/watch?v=Z6mlvrYEYnk (2013)

Books

• Peter Christen, Data Matching: http://www.springer.com/us/book/9783642311635
• Thomas N. Herzog, Fritz J. Scheuren and William E. Winkler, Data Quality and Record Linkage Techniques: http://www.springer.com/us/book/9780387695020

Free software

(last updated, stars)

Python

• Dedupe: https://github.com/dedupeio/dedupe (2017, 1,413)
• Python Record Linkage Toolkit: https://github.com/J535D165/recordlinkage (2017, 92)
• RLTK: https://github.com/usc-isi-i2/rltk/ (2017, 13)
• Febrl: https://sourceforge.net/projects/febrl/ (2013)
• ebLink for Python: https://github.com/aldengolab/graphical-record-linkage (2016, 4) (Python wrapper for ebLink)
• anonlink: anonymous linkage using cryptographic linkage keys https://github.com/n1analytics/anonlink (2018, 3)
• Entity Embed: scalable Record Linkage using Approximate Nearest Neighbors: https://github.com/vintasoftware/entity-embed/ (2021, 16)

Java

• Duke: https://github.com/larsga/Duke (2016, 469)
• JedAI: https://github.com/scify/JedAIToolkit (2017, 12)
• LSHDB: https://github.com/dimkar121/LSHDB (2017, 9)
• FRIL: http://fril.sourceforge.net/ (2011) (Windows binary available)

R

• RecordLinkage: https://r-forge.r-project.org/projects/recordlinkage/ https://cran.r-project.org/web/packages/RecordLinkage/index.html
• fastLink: https://github.com/kosukeimai/fastLink (2017, 47)
• ebLink: https://github.com/resteorts/ebLink (2016, 6)

Spark

• splink: https://github.com/moj-analytical-services/splink (2020, 9)

Other

• OpenRecLink: http://reclink.sourceforge.net/ (2016)
  • C++ with GUI
• Registry Plus™ Link Plus: https://www.cdc.gov/cancer/npcr/tools/registryplus/lp_tech_info.htm (2007)

Commercial software and solutions

• Data Ladder DataMatch: https://dataladder.com/
• Dedupe: https://dedupe.io/ (freemium frontend for Dedupe Python library)
• LinkageWiz: http://www.linkagewiz.net/
• WinPure Clean and Match: http://www.winpure.com/cleanmatch.html
• Reifier: http://nubetech.co/

For SAS

• (free but requires SAS) The Link King: http://www.the-link-king.com/

Data Cleaning

Name Parsers

Python

• probablepeople: https://github.com/datamade/probablepeople (2017, 204) python nameparser library (but purely syntax based)
• Name Parser: https://github.com/derek73/python-nameparser (2017, 232)
• NameCleaver: https://github.com/sunlightlabs/name-cleaver/ (2019, 111)
• nominally: https://github.com/vaneseltine/nominally (2019, 1)

JavaScript

• parse-full-name: https://github.com/dschnelldavis/parse-full-name (2017, 18)

Papers

• US Census Bureau Data Linkage research: https://www.census.gov/srd/csrm/RecordLinkage.html
• Learnable Similarity Functions and their Application to Record Linkage and Clustering (via)
• [https://www.microsoft.com/en-us/research/publication/improving-entity-resolution-with-global-constraints/ (via)
• https://arxiv.org/abs/1312.4645 (via)
• http://people.cs.umass.edu/~mwick/MikeWeb/Publications_files/wick09entity.pdf (via)
• http://homes.cs.washington.edu/~pedrod/papers/mrdm04.pdf (via)
• https://www.researchgate.net/publication/318874102_Clink_-_A_Novel_Record_Linkage_Methodology_based_on_Graph_Interactions
• Using a Probabilistic Model to Assist Merging of Large-scale Administrative Records (via)

Organizations

• US Census Burea - Center for Statistical Research and Methodolgy - Record Linkage: https://www.census.gov/srd/csrm/RecordLinkage.html
• Stanford Entity Resolution Framework: http://infolab.stanford.edu/serf/
• ANU Data Mining and Matching Group https://dmm.anu.edu.au/ https://web.archive.org/web/20160515215747/datamining.anu.edu.au/projects/linkage.html (Archive.org link)

Misc

• DuDe (framework for comparing record linkage results): https://hpi.de/naumann/projects/data-quality-and-cleansing/dude-duplicate-detection.html
• Datasets to use for evaluating deduplication software: https://hpi.de/naumann/projects/repeatability/datasets.html
• https://www2.vrdc.cornell.edu/news/3/20050420-Record%20Linkage%20Software.pdf

To Do

• list compatible data sources for software (CSV, SQL DB, JSON, data frame, etc...)
• GUI or not?
• list algorithms and techniques for softare (deterministic, probabalistic, graph, etc...)

Suggestions / contributions welcome! I am not an expert on record linkage, this is simply a list of things I've found when working on a difficult deduplication problem for Thicket.io.