Duplicate Record Resolver — Free Fuzzy Matching Tool

This tool finds near-duplicate records — names spelled slightly differently, the same person entered twice — using Levenshtein fuzzy matching. Levenshtein distance counts the smallest number of single-character edits (insertions, deletions, substitutions) that turn one string into another. Similarity is that distance scaled against the longer string. You set the threshold: pairs scoring at or above it are flagged for your review. The tool suggests; it never merges. Merging is a decision for a human who knows whether Anna Garcia and Ana Garcia are one person or two sisters.

How to use it

Paste the records, one per line.
Set the similarity threshold — 80 is a good starting point.
Click Find duplicates and read the pairs, highest similarity first.
Check each pair against your source data.
Merge only the pairs you have verified; keep a log of what you merged.

Worked example

A participant list contains Ana Garcia, Anna Garcia, Ana García, Bo Chen, and Bo Chan. At threshold 80, the tool flags Ana Garcia with Anna Garcia (91 percent) and Bo Chen with Bo Chan, while Ana García scores slightly lower because of the accented character. Checking the signup emails shows Anna Garcia is a typo of Ana Garcia, but Bo Chen and Bo Chan are different people who share a lab. One merge is correct, one would have been a serious error. That is why every flagged pair needs a human check, and why the tool reports similarity instead of deciding.

Honest limits

String similarity knows nothing about people. Twins, family members, and common names will score high while genuinely different spellings of one name — a married name, a transliteration — score low and escape. Accents, hyphens, and name order all affect the score, so normalize case and accents first with the Dirty String Standardizer for better results. Very long lists also get slow, because every record is compared with every other. For lists above a few hundred rows, block first: compare only within shared initials or shared groups.

Who it is for

This tool is for researchers deduplicating participant lists, event organizers cleaning registrations, and analysts merging datasets from two sources. It is a review aid, not an automatic merger. Its output belongs in your cleaning log: which pairs were flagged, which were merged, and what evidence justified each merge.

Going further

A solid dedup protocol has four steps and a paper trail. First, normalize: trim spaces, unify case, and strip accents so the matcher compares like with like. Second, block: split the list into small groups that could plausibly match — same first initial, same department, same cohort — and compare only within blocks, which also keeps the comparison count manageable. Third, review: a second person, or at least a second pass on a different day, checks every flagged pair against an independent field such as email or student id, because names alone are weak identifiers. Fourth, log: record each merge with the two original strings, the similarity score, the confirming field, and the date. Never merge on a single field for consequential records such as grades or consent: two different people merged into one record is a far worse error than one person left as two records, because the first corrupts analysis while the second only inflates a count you can explain. Prevention beats cleanup for the next round of data collection. Add a stable identifier to your forms — a student number, an email address, or a generated participant code — and validate it at entry, so the same person cannot register twice under two spellings. Where identifiers are impossible for privacy reasons, collect two quasi-identifiers, such as birth month plus initials, that let a future dedup check confirm matches without identifying anyone alone. The hour spent designing entry validation saves many hours of fuzzy matching later, and it prevents the worst outcome of all: merging two real people because their names looked alike.

Embed this tool on your website — free

Copy the code below to add this working tool to your own page. A small credit link stays with it — that’s all we ask.