CLEFeHealth 2014 Normalization of Information Extraction Challenge using Multi-model Method

This work focuses on making clinical documents easier to understand for patients and clinical workers. Normalization values of ten attributes have been predicted by the multi-model method which alternatively uses rule based methods and machine learning methods to solve different attribute problems. Information of text structure, lexical, and grammatical features are used to achieve overall average accuracy 0.787 and 0.849 on training data with run 1 and run 2, respectively. The UMLS CUI tool MetaMap is used to search for CUI category and CRFsuite package is adopted for machine learning method. In this paper, Run 1 is the official method and run 2 is considered as the supplement. Our system achieves overall average accuracy 0.793 on testing data with run 1 methods.