Entity linking for biomedical literature.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Computational Biology

Background:

• Entity Linking (EL) connects document mentions to knowledge base entities.
• EL is understudied in life sciences, hindering efficient computational model building.
• Existing news-trained EL methods perform poorly on biomedical data.

Purpose of the Study:

• To develop an unsupervised approach for EL in the life science domain.
• To enable efficient computational modeling of biological processes.
• To establish a benchmark for EL in biomedical literature.

Main Methods:

• Proposed a novel unsupervised collective inference approach.
• Leveraged ontology semantic information and structures for similarity computation and entity ranking.
• Applied to link entities from biomedical literature to 300 ontologies.

Main Results:

• Outperformed state-of-the-art supervised EL methods by 9% absolute linking accuracy.
• Achieved superior performance without manual annotation, unlike supervised methods requiring 15,000 annotations.
• Established a benchmark for the EL task in the life science domain.

Conclusions:

• The unsupervised collective inference approach effectively addresses EL in the life sciences.
• Demonstrated superior performance compared to heavily annotated supervised methods.
• Encourages further research in NLP and bioinformatics for this untapped domain.