作者: Mihai Surdeanu , Ramesh Nallapati , Julie Tibshirani , Christopher D. Manning
DOI:
关键词:
摘要: Distant supervision for relation extraction (RE) -- gathering training data by aligning a database of facts with text is an efficient approach to scale RE thousands different relations. However, this introduces challenging learning scenario where the expressed pair entities found in sentence unknown. For example, containing Balzac and France may express BornIn or Died, unknown relation, no at all. Because this, traditional supervised learning, which assumes that each example explicitly mapped label, not appropriate. We propose novel multi-instance multi-label RE, jointly models all instances their labels using graphical model latent variables. Our performs competitively on two difficult domains.