Word embeddings and external resources for answer processing in biomedical factoid question answering.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Biomedical question answering (QA) remains a significant challenge, with current deep learning models limited by the scarcity of large-scale biomedical QA datasets.
• Existing domain-independent QA methods struggle to adapt to the complexities of biomedical text.

Purpose of the Study:

• To develop and evaluate a novel, unsupervised machine learning approach for biomedical QA that leverages word embeddings.
• To improve the accuracy and efficiency of identifying and ranking candidate answers in biomedical texts.

Main Methods:

• An unsupervised, machine learning-based answer processing approach utilizing neural networks and word embeddings.
• Combining general and biomedical tools to identify candidate answers from passages.
• Representing candidates using features from external biomedical resources, textual sources, and word embeddings.
• Ranking candidates via a binary classification model trained on BioASQ challenge data.

Main Results:

• The integration of word embeddings alongside other features significantly enhances the performance of biomedical QA.
• Employing multiple annotators improves the accuracy of answer identification within passages.
• The proposed approach achieved competitive results in the BioASQ challenges of 2017 and 2018.

Conclusions:

• Word embeddings are a valuable component for improving unsupervised biomedical QA systems.
• A multi-annotator strategy enhances answer extraction accuracy.
• The developed approach demonstrates effectiveness and competitiveness in real-world biomedical QA tasks.