extract-entities_skill

Overview

This skill extracts named entities, topics, and relationships from free-form or structured text and returns a consistent JSON representation. It converts paragraphs, lists, and text fields into people, organizations, locations, topics, dates, key concepts, and relationship triples. The output is designed for downstream indexing, knowledge graph construction, and content linking.

How this skill works

The skill inspects text to identify clear entity mentions and normalizes them for consistency. It classifies entities into categories (people, organizations, locations, topics, dates, key concepts) and emits relationship triples (subject, predicate, object) for explicit or strongly implied links. It favors precision over recall, resolves pronouns only when antecedents are unambiguous, and avoids inventing entities not present in the source.

When to use it

• Extract structured entities from documents for search indexing or metadata tagging
• Generate relationship triples for knowledge graph building or entity linking
• Normalize dates and entity names across a corpus for analytics
• Detect collaborations, affiliations, or temporal orderings in text
• Preprocess text for summarization, recommendation, or pattern detection

Best practices

• Provide complete or well-scoped text snippets to reduce ambiguity
• Use the entity_types parameter to limit extraction to relevant categories
• Enable include_confidence to aid downstream filtering when needed
• Prefer explicit mentions over relying on implied references to avoid errors
• Review and reconcile conflicting relationships before automatic ingestion

Example use cases

• Indexing news articles: extract people, organizations, locations, and topics for faceted search
• Knowledge graph population: convert research papers into entities and affiliation triples
• Compliance review: find dates, parties, and contractual relationships in documents
• Content linking: identify related topics and authors across a document collection
• Analytics pipelines: normalize entity names and dates for cross-document trend analysis