Jedi: Joint Entity And Relation Detection Using Type Inference

FREEBASE contains entities and relation information but is highly incomplete. Relevant information is ubiquitous in web text, but extraction deems challenging. We present JEDI, an automated system to jointly extract typed named entities and FREEBASE relations using dependency pattern from text. An innovative method for constraint solving on entity types of multiple relations is used to disambiguate pattern. The high precision in the evaluation supports our claim that we can detect entities and relations together, alleviating the need to train a custom classifier for an entity type(1).