What is Entity Resolution?
Also called Record Linkage, Entity Matching.
Entity resolution is the process of determining that different records or mentions refer to the same real world thing, such as one customer appearing under three spellings across systems. It merges or links those references into a single identity so that facts about the entity accumulate in one place instead of fragmenting.
The problem arises whenever data comes from more than one source. A person may appear with a maiden name in one system, an abbreviated first name in another, and a typo in a third, while two genuinely different people may share a common name. Resolution must decide which pairs of records refer to the same entity.
Typical pipelines have three stages. Blocking groups records into candidate sets so that not every pair is compared. Matching scores candidate pairs using string similarity, embeddings, or learned models. Clustering then assigns records to entities, which must handle transitivity carefully, since a matching b and b matching c does not guarantee that a matches c.
Errors have asymmetric consequences. Merging two distinct entities creates a record mixing their attributes that is difficult to unpick later, and it can expose one person's information under another's identity. Failing to merge duplicates is less harmful and more recoverable, so conservative thresholds with human review of borderline pairs are common.
Entity resolution is foundational for knowledge graphs and for agent memory. A graph in which one company occupies four nodes cannot answer aggregate questions about it, and a memory store treating the same user as several people will recall preferences inconsistently. Resolution decisions are best stored as links rather than destructive merges.
Key points
- Decides when different records describe the same thing.
- Blocking, matching, and clustering are the usual stages.
- Wrong merges are harder to undo than missed merges.
- Essential for knowledge graphs and memory consistency.
- Store decisions as links, not destructive merges.
In practice
A customer appears as Acme Corp in the billing system, ACME Corporation in the support tool, and acme.com in web analytics. Entity resolution links the three to one organization identifier. A question about total spend and open tickets for that customer now returns one combined answer rather than three partial ones filed under different names.