External features enriched model for biomedical question answering.

Area of Science:

• Natural Language Processing
• Biomedical Informatics
• Artificial Intelligence

Background:

• Biomedical question answering (QA) is crucial for accessing healthcare information.
• Current neural network and pre-trained language model approaches show promise but have limitations due to biomedical corpus characteristics and small datasets.
• Significant room for improvement exists in biomedical QA performance.

Purpose of the Study:

• To enhance biomedical QA performance by incorporating external features.
• To address limitations in current models for the biomedical domain.
• To improve the accuracy and efficiency of healthcare information retrieval.

Main Methods:

• Proposed a novel framework to extract external features, including part-of-speech tagging and named-entity recognition.
• Fused these external features with text representations from pre-trained language models.
• Evaluated the framework on the BioASQ dataset for factoid question answering tasks.

Main Results:

• Achieved overall improvement across all three key metrics on BioASQ 6b, 7b, and 8b factoid QA tasks.
• Demonstrated the effectiveness of the external feature-enriched framework.
• Validated the positive impact of lexical and syntactic features on model performance.

Conclusions:

• The proposed framework effectively enhances biomedical QA.
• External lexical and syntactic features significantly improve the performance of pre-trained language models in the biomedical domain.
• This approach offers a promising direction for advancing biomedical question answering systems.