Entity resolution is the process of identifying records that refer to the same real-world entity and merging them into one canonical record. In commercial real estate it links a tenant, borrower, or property named differently across leases, rent rolls, and offering memoranda so a portfolio counts each entity once and reports on it accurately.
How Entity Resolution Works
Entity resolution works by comparing records, scoring how likely each pair refers to the same entity, then clustering matches into one canonical record. As Semantic Visions frames it, the task also called record linkage or deduplication identifies and merges data points that refer to the same real-world entity. Record linkage matches across datasets; deduplication matches within one.
The pipeline runs in stages. Blocking groups candidate records so the system does not compare every pair, which would be quadratic. A matching model then scores each candidate pair on fields like name, address, and identifier. Clustering resolves the pairwise scores into groups, and canonicalization merges each cluster into a single representation carrying the most complete information.
Accuracy is scored against a human-labeled key with precision, recall, and F1. In a Scientific Reports study on benchmark bibliographic data, F1 reached about 98% on Scopus and Google Scholar, and 100% on a clean Web of Science set. On messier real-world data, open tools diverge: Dedupe reports 99.4 to 100% recall but 71.6 to 73.7% precision, while Splink reports 88.6 to 94.1% precision, per a Tilores comparison.
Why Entity Resolution Matters
Entity resolution matters because unresolved duplicates corrupt every portfolio total that counts entities. Two spellings of one tenant inflate the tenant count and understate concentration risk. A borrower entered three ways across a loan book hides true exposure. In a document context, extraction produces clean fields that still name the same party inconsistently, so resolution is the step that makes cross-document rollups valid.
The failure is silent and expensive. A rent roll that treats "ABC Coffee LLC" and "ABC Coffee, L.L.C." as two tenants overstates tenant diversification, so a concentration limit that should trip does not. Because entity matching precision on real data can sit in the low 70s for some tools, a resolution step without human review of low-confidence merges risks either splitting one tenant into many or collapsing two distinct parties into one.
Example
Entity resolution is clearest when one tenant appears three ways across a portfolio and a naive count double-reports it. An analyst resolves tenant identity before computing concentration.
Record | Source document | Raw tenant name | Suite | Resolved entity |
1 | Lease A | ABC Coffee LLC | 210 | Entity 1 |
2 | Rent roll | ABC Coffee, L.L.C. | 210 | Entity 1 |
3 | Estoppel | ABC Cofee LLC | 210 | Entity 1 |
4 | Lease B | Bright Foods Inc | 305 | Entity 2 |
Before resolution, a naive distinct-name count returns four tenants, and ABC Coffee looks like 3 separate small tenants. After resolution, records 1 through 3 collapse to one entity on matching name and suite, so the true count is two tenants. If ABC Coffee pays 40,000 annually and Bright Foods pays 60,000, resolved tenant concentration for ABC Coffee is 40,000 / 100,000, or 40%, a figure the unresolved data hid entirely.
Variations and Edge Cases
Entity resolution behaves differently as the match signal weakens. The cases below are where a single field is not enough.
Case | Behavior |
Deduplication | Matches within one dataset, such as one rent roll |
Record linkage | Matches across datasets, such as lease to loan file |
Weak identifiers | Common names with no address force reliance on context |
Legal vs DBA | One entity trades under several registered names |
Over-merging | Two distinct parties share a name and get collapsed |
Scale | Blocking is required; pairwise comparison is quadratic |
Entity Resolution vs Data Normalization
Entity resolution is often confused with data normalization, but they solve different problems. Data normalization standardizes the format of a single value, such as rewriting a date to ISO 8601 or a rent to annual. Entity resolution decides whether two records point to the same real-world thing and merges them.
Normalization operates on one value in one record. Resolution operates across records to cluster identities. They run together: normalizing tenant names to a canonical case and punctuation makes the match easier, but even perfectly normalized records still need resolution to decide that "ABC Coffee" in a lease and "ABC Coffee" in a loan file are one borrower rather than two.
Frequently Asked Questions
What is entity resolution in commercial real estate?Entity resolution is the process of identifying records that refer to the same tenant, borrower, or property and merging them into one canonical record. It links parties named differently across leases, rent rolls, and loan files so a portfolio counts each entity once and measures concentration accurately.
What is the difference between entity resolution and deduplication?Deduplication is one form of entity resolution that matches records within a single dataset, such as one rent roll. Record linkage matches records across datasets, such as a lease against a loan file. Both identify records referring to the same entity and merge them into a canonical representation.
How accurate is entity resolution?On clean benchmark data, F1 can reach roughly 98% to 100%, per a Scientific Reports study. On messy real-world data, precision varies widely by tool, from the low 70s for Dedupe to the high 80s and low 90s for Splink, so low-confidence merges are reviewed before totals are computed.
Related Terms
Data Normalization
Named Entity Recognition
Schema Mapping
Rent Roll
Source Citation