SCINOBO: a novel system classifying scholarly communication in a dynamically constructed hierarchical...

Area of Science:

• Scientific publication classification
• Field-of-Science taxonomy development
• Research evaluation methodologies

Background:

• Classifying scientific publications is vital for applications like search engines and science monitoring.
• Existing classification schemes are often domain-specific, lack granularity, and require manual updates, struggling to keep pace with evolving research.
• Accurate classification supports stakeholders in organizing literature, calculating impact, and informing policy.

Purpose of the Study:

• To propose a novel hybrid approach for dynamically constructing a Field-of-Science taxonomy.
• To develop automatic classifiers for publication-level Field-of-Science assignment.
• To extend the OECD/FORD classification to a 5-level hierarchy for fine-grained analysis.

Main Methods:

• A hybrid approach combining metadata, bibliometric information, Neural Topic Modeling, and Community Detection.
• Dynamic construction of a Field-of-Science taxonomy based on OECD/FORD classification.
• Extension of a 3-level taxonomy to levels 4 and 5 using AI techniques, evaluated on Energy and Artificial Intelligence domains.

Main Results:

• Successfully generated an automatically constructed Field-of-Science taxonomy.
• The extended taxonomy (levels 4 and 5) demonstrates coherence and coverage in case studies.
• The proposed taxonomy captures the dynamics and emerging developments within research areas.

Conclusions:

• The novel hybrid approach effectively creates a dynamic and granular Field-of-Science taxonomy.
• This automated method addresses limitations of existing schemes, facilitating improved research evaluation and policy-making.
• The AI-driven taxonomy accurately reflects the evolving landscape of scientific research.