Procedural Knowledge Extraction on Medical Documents

We propose a procedural knowledge extraction on medical abstracts including procedural knowledge modeling. Procedural knowledge in medical documents consists of two main sections, purpose and solution sections. The purpose and solution section is considered as combinations of one or more unit processes which are triples of Target, Action, and Method. In order to extract procedural knowledge in those sections, we applied machine learning approaches like CRFs and SVMs, and utili zed part-of-speech tagging, parsing, predicate-argument structuring, and terminology mapping information as their features. In the end, we achieved 62%-82% F-scores in the internal steps.