Damerau–Levenshtein distance measures how many character edits—insertions, deletions, substitutions, and transpositions—transform one string into another. Larger distances indicate less similarity. section 3.4.3
In customer deduplication, it supports fuzzy comparison when names may contain typing differences. Candidate pairs can first be narrowed by postcode to reduce comparisons and balance accuracy against performance. section 3.4.3
A practical matching workflow is:
- Block records by postcode to create candidate links. section 3.4.3
- Compare first names and surnames with Damerau–Levenshtein thresholds, while matching postcodes exactly. section 3.4.3
- Keep pairs that satisfy all comparison criteria, then link their customer IDs back to the customer dataset. section 3.4.3
Thresholded customer matching
A recordlinkage comparison can use Damerau–Levenshtein for first_name and surname with threshold 0.85, alongside an exact postcode comparison. Matching pairs then become linked duplicate records.
