Fine-grained clinical outcome extraction and polarity classification.

Clinical outcome information is helpful for clinicians to understand the effect of a given intervention. Generally speaking, the outcome generated by an intervention has several aspects, and each aspect has different polarity. In this work we adopt structure learning algorithm to extract fine-grained outcome information and then determine the polarity of each aspect by trained classifier. Word and POS features are integrated by structure learning algorithm. The performance is evaluated on our labeled dataset. Experimental results indicate that although POS information can improve the performance of fine-grained outcome extraction it should be leveraged carefully.