Entity resolution
Figuring out that 'Apple Inc.', 'AAPL', and 'Apple Computer Company' all refer to the same organisation, so data about that entity can be unified.
The Simple Version
Figuring out that 'Apple Inc.', 'AAPL', and 'Apple Computer Company' all refer to the same organisation, so data about that entity can be unified.
Detailed Explanation
Entity resolution (also called record linkage, deduplication, or entity matching) is foundational to knowledge graph construction, master data management, and AI retrieval accuracy. Techniques range from rule-based matching (identical tax ID, same name) through probabilistic matching (probabilistic record linkage) to ML-based approaches (embedding similarity, Siamese networks). In AI retrieval contexts, poor entity resolution causes the same entity to be indexed multiple times under different representations, degrading both recall (missing some representations) and precision (returning duplicate results). Entity clarity in structured data (clear, consistent, unique entity identifiers) reduces the burden on entity resolution.
Key Characteristics
- Determines whether multiple data references denote the same real-world entity
- Methods: rule-based, probabilistic, and ML-based (embedding similarity)
- Foundational for knowledge graph quality and master data management
- Poor entity resolution degrades AI retrieval precision and recall
Why It Matters
Retailers building product knowledge graphs and enterprises integrating customer data across systems need entity resolution capabilities to create the consistent, deduplicated data that AI retrieval systems require for accurate performance.
Real-World Analogy
Like a post office that must determine that 'J. Smith, 12 Main St' and 'John Smith, No. 12 Main Street' are letters for the same person, combining reference variants to ensure correct delivery.
Common Misconceptions
- Entity resolution is a one-time data cleansing task, new data continuously introduces entity resolution challenges; it requires ongoing processes rather than a one-time project.
- Fuzzy string matching solves entity resolution, string similarity is one signal; robust entity resolution requires multiple evidence types including structured attributes, relationships, and context.