Leveraging word embeddings and medical entity extraction for biomedical dataset retrieval using unstructured texts

Area of Science:

• Biomedical Informatics
• Data Science
• Information Retrieval

Background:

• Open data initiatives have increased the volume of accessible biomedical datasets.
• The biomedical and healthCAre Data Discovery Index Ecosystem (bioCADDIE) portal aggregates these datasets.
• Efficient retrieval of relevant biomedical datasets is a growing challenge due to data volume.

Purpose of the Study:

• To develop and evaluate an advanced information retrieval (IR) system for the bioCADDIE Dataset Retrieval Challenge.
• To enhance the accuracy and efficiency of searching and retrieving biomedical datasets.

Main Methods:

• Implemented a state-of-the-art IR model utilizing dataset titles and descriptions.
• Incorporated medical named entity extraction and query expansion with deep learning-based word embeddings.
• Employed a re-ranking strategy to optimize retrieval performance.

Main Results:

• The proposed IR system outperformed 11 baseline systems in empirical experiments.
• Achieved superior performance in terms of inference Average Precision and inference normalized Discounted Cumulative Gain.
• Demonstrated the system's viability for effective biomedical dataset retrieval.

Conclusions:

• The developed IR system significantly improves the retrieval of relevant biomedical datasets.
• The system's advanced techniques offer a viable solution for navigating large-scale biomedical data repositories.
• Facilitates accelerated scientific discovery by enhancing data accessibility and reuse.