Data analyst · Brisbane, Australia

Kyle Haynes

Data Integration team, Queensland Government Statistician’s Office (QGSO)

I work on data linkage — joining records across data sets that don’t share a clean key — which pulls in a fairly specific mix of theory: graph theory for the connections between records, and string matching for the fuzzy, almost-but-not-quite-identical values that connect them.

  • data linkage
  • graph theory
  • string matching
  • R & data.table
  • spatial / addresses

What I work on

Most data linkage problems reduce to the same question: do these two records describe the same real-world thing? Answering that well means treating records and shared values as a graph — connected components, not pairwise rules — and being deliberate about how string similarity is measured, since names, addresses, and identifiers are rarely typed the same way twice.

  • Data linkage — matching and deduplicating records across sources with no reliable shared key.
  • Graph theory — connected components, clustering, and traversal as tools for resolving messy links.
  • String matching — fuzzy and approximate matching for names, addresses, and free text.

Active projects

gnafr

R package for fuzzy matching Australian addresses against GNAF.
Visit doco →
Visit repo →

fast.string

Fast string processing utilities used as building blocks for matching and cleaning text fields.
Visit doco →
Visit repo →

graphfast

Limited, though, fast graphing methods
Read the post →
Visit repo →

jsdiffr

Diffs in R!
Read the post →
Visit repo →