PPR-SSM: personalized PageRank and semantic similarity measures for entity linking

Area of Science:

• Biomedical informatics
• Text mining
• Bioinformatics

Background:

• Biomedical literature requires controlled vocabularies for consistent terminology.
• Rapidly advancing fields like drug development introduce ambiguity.
• Entity linking connects literature mentions to knowledge base concepts.

Purpose of the Study:

• To develop an improved entity linking method for biomedical literature.
• To address ambiguity by incorporating domain-specific semantic information.
• To enhance the accuracy of linking named entities to biomedical ontologies.

Main Methods:

• Proposed Personalized PageRank with Semantic Similarity Measures (PPR-SSM).
• Utilized domain-specific ontologies to create concept graphs.
• Weighted concept graph edges using semantic similarity measures (SSM).

Main Results:

• PPR-SSM effectively links entities like chemical compounds, phenotypes, and gene products.
• The method demonstrated improved accuracy in entity linking.
• Achieved a 0.1385 improvement in chemical compound entity linking accuracy compared to non-SSM methods.

Conclusions:

• PPR-SSM outperforms state-of-the-art entity linking methods.
• The graph-based approach leverages semantic information from ontologies.
• PPR-SSM does not require training data, making it broadly applicable.