An entity type is the category an entity belongs to: a person, an organisation, a location, a product, an event, and so on. Every system that works with entities uses a set of types to label them, from entity extraction APIs to Schema.org markup and knowledge graphs. Knowing the type tells a machine what kind of thing it is dealing with and which attributes to expect.
How it works
Each system has its own list of types, so the same thing can carry different labels depending on the tool:
- NLP APIs. Google Cloud Natural Language returns a fixed set of types including PERSON, LOCATION, ORGANIZATION, EVENT, WORK_OF_ART, CONSUMER_GOOD, OTHER, plus special types such as PHONE_NUMBER, ADDRESS, DATE, NUMBER and PRICE. Other APIs (Amazon Comprehend, IBM Watson NLU) use different lists.
- Research benchmarks. The CoNLL-2003 named entity recognition task used four types: persons, organisations, locations and miscellaneous names.
- Schema.org. Types are arranged in a hierarchy (Thing, then Organization, then LocalBusiness, and so on), and each type comes with its own properties.
- Knowledge graphs. In Wikidata, types are expressed through “instance of” and “subclass of” statements, which makes them part of an ontology.
The type matters because attributes depend on it. In MLforSEO’s EAV framing, each entity type has its own attributes and each attribute its own variables: a product has price and size, a person has a role and an employer.
How it’s used in SEO and marketing
- Search intent classification: MLforSEO’s semantic parsing approach treats entity types as signals, so CONSUMER_GOOD and PRICE push a query towards transactional and ORGANIZATION towards navigational.
- Entity inventories: the Entity SEO course builds a brand’s entity catalogue across types such as organisation, leadership, products and services, locations, events and concepts.
- Schema audits: the Schema Entity Audit Template scores each entity type on presence and quality.
- Generating synthetic queries: a Semantic Keyword Research lab generates different query patterns per type, for example “reviews” and “alternatives” for products, “biography” for people.
Example
Run “Acme launches the Nimbus 3 trainer at £120 in Manchester” through Google’s API and you get Acme as ORGANIZATION, Nimbus 3 as CONSUMER_GOOD, £120 as PRICE and Manchester as LOCATION. That single pass tells you the sentence is commercial, names a product and has a local angle.
Limitations and things to watch
- Types don’t line up across tools. The MLforSEO comparison of NLP APIs found Google labelling concepts such as machine learning as OTHER, while Amazon Comprehend returned them only as key phrases.
- Broad types such as OTHER hide a lot of variety, so check samples before you build rules on them.
- An NLP type is a guess from the text; the type you declare in schema is a claim you control. Keep the two consistent for your own brand.
Learn it on MLforSEO
