Automatically extracting information needs from Ad Hoc clinical questions.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Automating information extraction from clinical questions is crucial for advancing medical question answering.
• Ad hoc clinical questions present unique challenges for automated processing.

Purpose of the Study:

• To explore supervised machine learning for classifying ad hoc clinical questions into general topics.
• To evaluate various methods for automatically extracting keywords from clinical questions.

Main Methods:

• Supervised machine learning algorithms were employed for topic classification.
• Different keyword extraction techniques were assessed.
• Methods were validated on a dataset of 4,654 clinical questions from the National Library of Medicine.

Main Results:

• The best performing system achieved an F-score of 76% for question-topic classification.
• An average F-score of 56% was obtained for keyword extraction from clinical questions.

Conclusions:

• Supervised machine learning demonstrates effectiveness in classifying clinical question topics.
• Keyword extraction from clinical questions remains a challenging task with room for improvement.