Knowledge-based citation reasoning for biomedical domain.

Area of Science:

• Biomedical Informatics
• Natural Language Processing
• Scholarly Communication

Background:

• Citation is crucial for navigating scientific literature, but the reasons for citation are often unclear.
• Existing academic search tools lack transparency in ranking motivations, hindering interpretation, especially in specialized fields like biomedicine.

Purpose of the Study:

• To develop a method for explaining citation motivations in biomedical research.
• To improve the transparency and interpretability of academic search results.

Main Methods:

• An encoder-decoder framework was developed, integrating curated biomedical knowledge.
• The model generates explanations of citation motivation in a structured bio-triplet format.
• Experiments involved annotating over 10,000 cancer-related citation relations from PubMed Central.

Main Results:

• The proposed framework outperformed existing sequence-generation models (BERT-style, GPT-style).
• Improvements were observed in precision, recall, and F1 scores for citation motivation generation.
• The model successfully generated bio-triplets grounded in established biomedical databases.

Conclusions:

• This work provides a novel approach to elucidating citation reasoning in biomedical literature.
• The developed framework enhances the utility of academic search by offering transparent citation explanations.
• Future work can extend this method to other scientific domains and explore deeper knowledge integration.