A knowledge-driven approach to biomedical document conceptualization

Area of Science:

• Biomedical Informatics
• Computational Biology
• Information Retrieval

Background:

• Biomedical document clustering typically relies on static domain knowledge.
• Existing methods lack flexibility in adapting to diverse user needs and ontologies.
• Effective conceptualization of biomedical documents is crucial for knowledge extraction.

Purpose of the Study:

• To develop a flexible framework for clustering biomedical documents based on user-specified ontologies.
• To enable users to leverage specific domain knowledge for more effective document conceptualization.
• To improve the representation of biomedical documents using key concepts and their relationships.

Main Methods:

• A flexible framework was developed to incorporate user-specified ontologies as knowledge bases.
• A key concept induction algorithm utilizing latent semantic analysis was implemented for document clustering.
• A corpus-related ontology generation algorithm was created to derive document conceptual structures.

Main Results:

• The proposed method outperformed five other clustering algorithms in key concept identification on two biomedical datasets.
• Achieved F-measure values ranged from 0.5294 to 0.7294 using MeSH and Gene Ontology (GO).
• Generated corpus-related ontologies demonstrated informative conceptual structures for both MeSH and GO.

Conclusions:

• The developed method allows users to define domain knowledge for enhanced biomedical document analysis.
• The approach effectively extracts key concepts and clusters documents with high precision.
• This facilitates a more nuanced understanding of biomedical document collections.