An Information Extraction Approach to Detecting Novelty of Biomedical Publications.

Area of Science:

• Biomedical Informatics
• Bibliometrics
• Scientific Communication

Background:

• Scientific novelty is crucial for research impact but is hard to define and quantify.
• Current methods often oversimplify novelty, failing to capture diverse contributions like new concepts or relationships.

Purpose of the Study:

• To develop a semantic measure of novelty based on new biomedical entities and relationships in research article conclusions.
• To classify articles into novelty categories: No Novelty, Entity-only Novelty, Relation-only Novelty, and Entity-Relation Novelty.
• To evaluate the correlation between these novelty types and research influence using citation counts and Journal Impact Factors (JIF).

Main Methods:

• Utilized transformer-based named entity recognition (NER) and relation extraction (RE) tools.
• Analyzed conclusion sections of research articles to identify novel biomedical entities and relationships.
• Categorized articles based on the presence and type of novelty detected.

Main Results:

• Articles categorized as Entity-Relation Novelty demonstrated the highest citation impact.
• Novelty in relationships showed a stronger alignment with high-impact journals compared to entity-only novelty.
• The proposed framework provides a scalable method for assessing scientific novelty.

Conclusions:

• A semantic approach to novelty assessment, distinguishing between entity and relation novelty, offers deeper insights into research impact.
• Entity-Relation Novelty is a key indicator of significant research influence.
• This framework can guide research evaluation and identify high-impact scientific contributions.