Result identification for biomedical abstracts using Conditional Random Fields

For biomedical research, the most important parts of an abstract are the result and conclusion sections. Some journals divide an abstract into several sections so that readers can easily identify those parts, but others do not. We propose a method that can automatically identify the result and conclusion sections of any biomedical abstracts by formulating this identification problem as a sequence labeling task. Three feature sets (Position, Named Entity, and Word Frequency) are employed with Conditional Random Fields (CRFs) as the underlying machine learning model. Experimental results show that the combination of our proposed feature sets can achieve F-measure, precision, and recall scores of 92.50%, 95.32% and 89.85%, respectively.