AI glossary · Advanced · Data engineering
Entity Resolution
Entity resolution (finding and merging records of the same entity) is the task of determining whether records in different datasets, with slightly different names or spellings, actually refer to the same person, company or product, and combining them into one.
About this term
- Level
- Advanced · Specialist terms that need the inner workings
- Category
- Data, ops & platforms
- Fields
- Data engineering
- Added
- 2026-10-09
How it works
First, preprocessing cleans up spelling differences in names, such as spacing, capitalization and abbreviations ("Co.", "Corporation"). Comparing every record with every other record would take too long, so "blocking" reduces the candidates to compare by splitting records into groups, for example by region or the first letter of the name. For each candidate pair, a score is computed for how similar the names, addresses and phone numbers are, and rules or a machine learning model decide whether they are the same entity. Records judged to be the same are grouped together to create a representative record.
Narrowing down candidates
Why it matters
If the same customer is counted multiple times, customer counts and sales analysis go wrong, and if the same company is scattered under several names, transaction risk can't be seen properly. It is a required step when combining data from multiple departments and systems to build a data warehouse or knowledge graph. In AI training data, it is closely related to deduplication, which keeps the same content from being included more than once. Approaches that use embeddings to find records with similar meanings, or have an LLM judge ambiguous pairs, are also used today.
Things to know
There is always a trade-off between wrongly merging records and missing records that should be merged. In fields like healthcare and finance, where wrong merges cause serious harm, a process is put in place for people to check ambiguous cases. Because the records contain personal information, personal data de-identification and access control also need to be considered. Data keeps coming in, so this is not a one-time task but one that must be updated continuously.
Example
A shopping site wants to combine online membership data, offline store memberships and customer service records to build a "complete picture of each customer." Even if the same customer signed up in-store with an English name and online with a Korean name, entity resolution groups them as one person if the phone number and address are similar. As a result, duplicate coupon mailings drop, and the recommendation system can make recommendations based on the customer's full purchase history.