Fuzzy Matching & Record Linkage: Normalising Names, Similarity Scores & Duplicates
Normalise messy company names, block candidate pairs, score them with bigram and token similarity, combine the evidence into match / review / non-match decisions, measure precision and recall, and merge duplicates into clusters and golden records, in a live spreadsheet that checks every cell.
The first lesson is free — no account needed.
About this course
A fuzzy matching and record linkage course that finds the duplicate customers in fictional company lists in a spreadsheet, for data analysts, CRM and master data staff, finance teams cleaning vendor files and data science students. You normalise messy company names with TRIM, LOWER, SUBSTITUTE and a list of legal suffixes, build blocking keys from first letter and postcode and count the comparisons they save, measure how many true matches exact and normalised matching find, score name pairs with character bigrams and the Jaccard and Dice similarity using MID and SEQUENCE, split names into words with TEXTSPLIT for token overlap, combine four pieces of evidence into a weighted score with match, review and non-match thresholds, test the matcher against 20 labelled pairs with precision, recall and F1, and merge duplicates into clusters and golden records with MINIFS. Fictional companies and data. The first lab is free.
Lessons
- Normalising Company Names: TRIM, LOWER, SUBSTITUTE and Legal SuffixesFree
- Blocking Keys: First Letter + Postcode to Limit Comparisons Pro
- Exact vs Normalised Matching: How Many True Matches Do You Find? Pro
- Character Bigram Similarity: Jaccard and Dice with MID and SEQUENCE Pro
- Token Overlap for Company Names: Word Jaccard and the Overlap Coefficient Pro
- Combined Match Score: Weights and Two Thresholds (Match, Review, Non-Match) Pro
- Precision and Recall of the Matcher against a Labelled Sample Pro
- De-duplication: Cluster IDs, the Golden Record and the Score Distribution Pro