Dual Triggered Correspondence Topic (DTCT)model for MeSH annotation.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Information Retrieval

Background:

• Accurate Medical Subject Headings (MeSH) annotation is crucial for effective biomedical literature retrieval and knowledge discovery.
• Existing methods may struggle to balance general topic relevance with specific keyword matching for MeSH annotation.

Purpose of the Study:

• To develop a novel Dual Triggered Correspondence Topic (DTCT) model for automated MeSH annotation.
• To enhance the accuracy and efficiency of MeSH annotation in biomedical research.

Main Methods:

• Developed a DTCT model where abstracts and titles describe MeSH terms, assuming shared latent topic factors.
• Integrated Okapi BM25 for discriminative feature selection to mitigate non-topical word influence.
• Incorporated prior word-MeSH relationship knowledge using the phi-coefficient to improve topic coherence.

Main Results:

• The DTCT model achieved an F-score of 0.62 on a large MEDLINE test set.
• Demonstrated a particular strength in recall rate for MeSH annotation.
• Showcased competitive performance within an integrated probabilistic framework without post-processing.

Conclusions:

• The DTCT model offers a powerful and effective approach for automatic MeSH annotation.
• The model successfully balances topic-based abstraction and specific word matching for improved accuracy.
• This method enhances biomedical information retrieval and knowledge discovery.