ABSTRACT
Relation extraction is always seen as a classification problem, many machine learning methods are attempted to solve this task. Roberts et al. (2008) used SVM to extract relations in their information extraction system CLEF, they compared the performance of relations within a sentence and across a sentence, the results showed that relation extraction across sentence is more difficult. Wang et al. (2009) annotated corpus for entities and relations based on SNOMED-CT by themselves and combined CRF and SVM together to extract the relations. The 2010 Informatics for Integrating Biology and the Bedside (i2b2)/Veteran’s Affairs (VA) challenge provided an opportunity for all participants to demonstrate their methods on relation extraction (Uzuner et al. 2011). In this challenge, DeBruijn et al. (2011) ranked high among all the submitted systems with an F-score of 0.7313, they found unlabeled data and syntactic dependency structures are useful to improve the performance. Grouin et al. (2010) imported linguistic patterns to improve their system and gained an F-score of 0.709.
