0 / 0
Relations extraction

Relations extraction

Watson Natural Language Processing Relations extraction encapsulates algorithms for extracting relations between two entity mentions. For example, in the text Lionel Messi plays for FC Barcelona. a relation extraction model may decide that the entities Lionel Messi and F.C. Barcelona are in a relationship with each other, and the relationship type is works for.

Capabilities

Use this model to detect relations between discovered entities.

The following table lists common relations types that are available out-of-the-box after you have run the entity models.

Table 1. Available common relation types between entities
Relation Description
affiliatedWith Exists between two entities that have an affiliation or are similarly connected.
basedIn Exists between an Organization and the place where it is mainly, only, or intrinsically located.
bornAt Exists between a Person and the place where they were born.
bornOn Exists between a Person and the Date or Time when they were born.
clientOf Exists between two entities when one is a direct business client of the other (that is, pays for certain services or products).
colleague Exists between two Persons who are part of the same Organization.
competitor Exists between two Organizations that are engaged in economic competition.
contactOf Relates contact information with an entity.
diedAt Exists between a Person and the place at which he, she, or it died.
diedOn Exists between a Person and the Date or Time on which he, she, or it died.
dissolvedOn Exists between an Organization or URL and the Date or Time when it was dissolved.
educatedAt Exists between a Person and the Organization at which he or she is or was educated.
employedBy Exists between two entities when one pays the other for certain work or services; monetary reward must be involved. In many circumstances, marking this relation requires world knowledge.
foundedOn Exists between an Organization or URL and the Date or Time on which it was founded.
founderOf Exists between a Person and a Facility, Organization, or URL that they founded.
locatedAt Exists between an entity and its location.
managerOf Exists between a Person and another entity such as a Person or Organization that he or she manages as his or her job.
memberOf Exists between an entity, such as a Person or Organization, and another entity to which he, she, or it belongs.
ownerOf Exists between an entity, such as a Person or Organization, and an entity that he, she, or it owns. The owner does not need to have permanent ownership of the entity for the relation to exist.
parentOf Exists between a Person and their children or stepchildren.
partner Exists between two Organizations that are engaged in economic cooperation.
partOf Exists between a smaller and a larger entity of the same type or related types in which the second entity subsumes the first. If the entities are both events, the first must occur within the time span of the second for the relation to be recognized.
partOfMany Exists between smaller and larger entities of the same type or related types in which the second entity, which must be plural, includes the first, which can be singular or plural.
populationOf Exists between a place and the number of people located there, or an organization and the number of members or employees it has.
measureOf This relation indicates the quantity of an entity or measure (height, weight, etc) of an entity.
relative Exists between two Persons who are relatives. To identify parents, children, siblings, and spouses, use the parentOf, siblingOf, and spouseOf relations.
residesIn Exists between a Person and a place where they live or previously lived.
shareholdersOf Exists between a Person or Organization, and an Organization of which the first entity is a shareholder.
siblingOf Exists between a Person and their sibling or stepsibling.
spokespersonFor Exists between a Person and an Facility, Organization, or Person that he or she represents.
spouseOf Exists between two Persons that are spouses.
subsidiaryOf Exists between two Organizations when the first is a subsidiary of the second.

In Runtime 22.2, relation extraction is provided as an analysis block, which depends on the Syntax analysis block and a entity mention extraction block. Starting with Runtime 23.1, relation extraction is provided as a workflow, which is directly run on the input text.

Relation extraction in Runtime 23.1

Workflow name

relations_transformer-workflow_multilingual_slate.153m.distilled

Supported languages The Relations Workflow is available for the following languages. For a list of the language codes and the corresponding language, see Language codes.

ar, de, en, es, fr, it, ja, ko, pt

Code sample

import watson_nlp

# Load the workflow model
relations_workflow = watson_nlp.load('relations_transformer-workflow_multilingual_slate.153m.distilled')

# Run the relation extraction workflow on the input text
relations = relations_workflow.run('Anna Smith is an engineer. Anna works at IBM.', language_code="en")
print(relations.get_relation_pairs_by_type())

Output of the code sample

{'employedBy': [(('Anna', 'Person'), ('IBM', 'Organization'))]}

Relation extraction in Runtime 22.2

Block name

relations_transformer_en_stock

Supported languages

The Relations extraction block is available for English only.

Dependencies on other blocks

The following block must run before you can run the relations_transformer_en_stock block:

  • syntax_izumo_en_stock

This must be followed by one of the following entity models on which the relations extraction block can build its results:

  • entity-mentions_rbr_en_stock
  • entity-mentions_bert_multi_stock

Code sample

import watson_nlp

# Load the models for English
syntax_model = watson_nlp.load('syntax_izumo_en_stock')
entity_mentions_model = watson_nlp.load('entity-mentions_bert_multi_stock')
relation_model = watson_nlp.load('relations_transformer_en_stock')

# Run the prerequisite models
syntax_prediction = syntax_model.run('Anna Smith is an engineer. Anna works at IBM.')
entity_mentions = entity_mentions_model.run(syntax_prediction)

# Run the relations model
relations_on_mentions = relation_model.run(syntax_prediction, mentions_prediction=entity_mentions)
print(relations_on_mentions.get_relation_pairs_by_type())

Output of the code sample:

{'employedBy': [(('Anna', 'Person'), ('IBM', 'Organization'))]}

Parent topic: Watson Natural Language Processing task catalog

Generative AI search and answer
These answers are generated by a large language model in watsonx.ai based on content from the product documentation. Learn more