Data Cleaning Exercises

Module 10

Entity resolution

Candidate pairs, exact versus fuzzy evidence, false merges and human review.

  1. Merge on evidence, review the rest

    Decide which records describe the same customer using strong evidence, apply a declared survivor rule, and route weak cases to human review. · 10 min

  2. Evidence for a merge

    Decide when records may be merged automatically, when they need review, and when they stay apart. · 9 min

  3. Human review queues

    Route uncertain matches to people with the evidence they need to decide. · 6 min

  4. Blocking candidate pairs

    Limit comparisons to plausible pairs with a cheap blocking key. · 7 min

Module 10 quiz

Practice questions: choose an answer and check it to see the explanation. Nothing here is timed or graded.

  1. 1. What is a false merge?
  2. 2. Two records have the same name but different emails and no shared id. What should happen?
  3. 3. Records A (email " Ann@X.com") and B ("ann@x.com") under a trim-and-lowercase email policy: same key?
  4. 4. What is blocking in entity resolution?
  5. 5. A merge rule matches on name only. What is the main risk?
  6. 6. Records have no email. The policy merges on normalized email. What happens to them?