Entity and relationship extraction converts a text document into structured graph knowledge. It identifies entities from a supplied type list, records each entity’s name, type, and description, then finds clearly related entity pairs and assigns relationship descriptions and strength scores. section 7.2.2
A practical extraction workflow:
- Choose entity types that match the dataset, including general categories such as people, organizations, and locations or domain-specific categories. section 7.1
- Send each text chunk with an extraction prompt, parse the model’s delimited output, and separate nodes from relationships. section 7.2.2
The Odyssey
The Odyssey supplies recurring entities such as Ulysses, people, gods, mystical weapons, and events. Repeated appearances of Ulysses across chunks support testing entity extraction and cross-chunk relationships. section 7.1
