When appraising published clinical research, medical doctors and researchers often need to know whether the clinical outcomes presented had statistical evidence. In this paper we present a study for the detection of expressions of such statistical evidence. An effective rule-based classifier has been developed that uses regular expressions and a list of negation phrases to automatically classify documents as either showing evidence of effect in the results or not. The classifier performed with an accuracy between 88% and 98% at 95% confidence intervals, and it also outperformed a set of baselines using bag-of-word features in several statistical classifiers. The rule-based system is written in Python and is available as open-source code.
|Name||Conferneces in research and practice in information technology|
|Publisher||Australian Computer Society|
|Workshop||Australasian Workshop on Health Informatics and Knowledge Management (5th : 2012)|
|Period||30/01/12 → 3/02/12|
- evidence-based medicine
- text classification