Knowledge-Guided Data Cleaning

A research survey connecting knowledge acquisition to reliable data-cleaning decisions.

Knowledge-Guided Data Cleaning: From Knowledge Acquisition to Reliable Decisions is a research survey covering 30 studies from 2007 to 2026. It examines how rules, facts, patterns, and feedback support data repairs, and how those decisions connect to reliable, task-oriented data preparation.

Diagram showing trusted knowledge guiding the repair of a conflicting ZIP and city record
Conceptual illustration: trusted evidence supports selecting a repair for a conflicting record.

Survey Focus

  • Knowledge acquisition: Where do cleaning rules, trusted facts, data patterns, and user feedback come from?
  • Repair selection: What evidence supports a particular correction to a record or attribute?
  • Reliability: When should a system apply a repair, revise its knowledge, or abstain and seek more evidence?
  • Data preparation: How do semantic interpretation, transformation, joins, and operation order affect the intended analytical result?

The survey distinguishes factual correctness, constraint consistency, and downstream task utility. Its take-home principles emphasize retaining verified knowledge, updating it when conditions change, seeking missing evidence, carrying knowledge across tasks, and checking each operation against the intended result.

Project Materials

The project website includes an interactive ZIP–city repair example and a searchable literature map organised by acquisition, repair, reliability, and preparation. Each study can be expanded for a concise explanation.

  • Survey paper: a seven-page overview of the literature and its connecting questions.

Project Website · Paper · GitHub

← Back to Projects