What This Tool Does
Use this browser-based tool to group alternate spellings, casing, and formatting within selected columns. It clusters values such as "USA", "U.S.A.", and "United States", then asks you to approve, edit, or reject a canonical value. It works within columns; it does not decide whether whole rows are duplicate records.
Similarity Methods
Normalized text compares values after trimming, collapsing whitespace, and lowercasing - it catches pure formatting differences. Fingerprint builds an OpenRefine-style key from the sorted, deduplicated words in a value, so "Smith, John" and "John Smith" land in the same cluster. Edit-distance uses a Levenshtein similarity ratio against your threshold to catch typos like "Recieved" vs "Received". Phonetic uses Soundex codes to catch values that sound alike but are spelled differently, like "Catherine" and "Kathryn". For normalized text, fingerprint, and phonetic matching, the threshold also merges clusters whose representative values are close enough under normalized-text similarity. Edit-distance uses the threshold for its Levenshtein comparisons.
How to Use the Value Clusterer & Standardizer
- Load a file: Drop a CSV, TSV, JSON, JSONL, XLSX, or XLS file, or click browse files. Delimited files use the first row as headers. JSON must be an array of records or an object containing an array of record objects.
- Configure clustering: Select columns, choose a similarity method, set the threshold, and choose Replace in column or Write to new column.
- Find clusters: Select one or more columns; clustering updates automatically whenever the source or clustering settings change.
- Review results: Clusters start as Awaiting review. Edit the proposed canonical value if needed, then click Approve mapping or Reject (leave values as-is). The summary shows how many cells would change.
Review and Choose the Output Column
Review cards can show Approved, Approved (edited), Rejected, or Awaiting review. Replace in column overwrites the selected source column for approved mappings. Write to new column creates <column> (standardized) and keeps the original column. Rows are not merged or deleted.
Export and Reapply Mappings
Click Export cleaned dataset (CSV) to download the result or Export mapping table (JSON) to save approved mappings. If clusters are still pending, the dataset export asks you to confirm; unreviewed values remain unchanged. To reuse a mapping, load the new dataset first, then click Import mapping to reapply. Imported matches are shown for review again before replacement.
File Limits and Scope
Browser processing accepts up to 20,000 rows. Files over 3,000 rows may take several seconds to cluster; files over 20,000 rows are rejected and should be split. Empty files and JSON without an array of record objects cannot be loaded.
Parsing, clustering, and export happen in your browser. This tool never merges or deletes rows - that's the job of the Fuzzy Duplicate Finder, which compares whole records instead of values within one column. It doesn't call any external service to infer canonical entities (no gazetteer or geocoding lookups); every proposed canonical value is derived deterministically from the values already in your file, most often the most frequent variant in a cluster. Boolean, country, state, casing, abbreviation, and spelling normalization aren't separate tools here - they're just clusters you review and approve like any other.
Frequently Asked Questions
How is this different from the Fuzzy Duplicate Finder?
The Fuzzy Duplicate Finder compares whole records to decide whether two rows represent the same entity. This tool works within a single column, grouping alternate spellings, casing, or formatting of the same value so you can standardize them. It never decides whether records are duplicates.
What similarity methods are available?
Normalized text, fingerprint, edit-distance, and phonetic. For normalized text, fingerprint, and phonetic matching, the similarity threshold can also merge clusters whose representative values are close enough under normalized-text similarity. Edit-distance uses the threshold for its Levenshtein comparisons.
Will this tool replace values automatically?
No. Every cluster starts as Awaiting review. Nothing is replaced until you approve that cluster's mapping, optionally after editing the canonical value.
Can I reuse a mapping on another file?
Yes. Export the mapping table as JSON, load the new dataset first, and use Import mapping to reapply. Matching entries are shown for review again before anything is replaced.