A new and efficient method based on syntactic dependency relations features for ad hoc clinical question classification

Abstract

Clinical question classification is an important and a challenging task for any clinical Question Answering (QA) system. It classifies questions into different semantic categories, which indicate the expected semantic type of answers. Indeed, the semantic category allows filtering out irrelevant answer candidates. Existing methods dealing with the problem of clinical question classification don't take into account the syntactic dependency relations in questions. Therefore, this may impact negatively the performance of the clinical question classification system. To overcome this drawback, we propose to incorporate the syntactic dependency relations as discriminative features for machine learning. To evaluate and illustrate the interest of our contribution, we conduct a comparative study using nine methods and two machine-learning algorithms: Naive Bayes and Support Vector Machine (SVM). The obtained results using 4654 clinical questions maintained by the National Library of Medicine (NLM) show that our proposed method is very efficient and outperforms greatly the others by the average F-score of 4.5% for Naive Bayes and 4.73% for SVM.