Resource and Response Type Classification for Consumer Health Question Answering.

Area of Science:

• Natural Language Processing
• Health Informatics
• Machine Learning

Background:

• Health question answering systems rely on accurate question type classification.
• Several modeling choices exist for this crucial initial step.
• Optimizing classification is key to improving health information retrieval.

Purpose of the Study:

• To evaluate the effectiveness of various modeling choices for health question type classification.
• To identify optimal embeddings and architectural hyperparameters for classifiers.
• To advance the performance of consumer health question classification.

Main Methods:

• Evaluated different sentence classification methods for consumer health questions.
• Assessed various embedding techniques and architectural hyperparameters.
• Utilized the GARD dataset for performance evaluation.
• Developed a new dataset of 2,882 annotated medical questions.

Main Results:

• Achieved a new best 5-fold accuracy of 85.3% on the GARD dataset.
• Demonstrated improved performance over previous methods.
• Identified effective modeling choices for health question classification.

Conclusions:

• Sentence classification methods can be effectively applied to consumer health question type classification.
• The study contributes valuable insights and a new dataset for this task.
• Optimized models enhance the accuracy of health question answering systems.