Research Ideas Discovery via Hierarchical Negative Correlation.

Area of Science:

• Information Science
• Computer Science
• Bibliometrics

Background:

• Discovering novel research ideas is crucial for scientific advancement.
• Keyword co-occurrence analysis is a common method, but struggles with diverse domain patterns.
• Existing link prediction methods may not capture the complexity of interdisciplinary research.

Purpose of the Study:

• To propose a novel link prediction framework for identifying new research avenues.
• To enhance the diversity of extracted patterns by addressing domain-specific author habits.
• To develop an efficient and computationally cost-effective method for research idea discovery.

Main Methods:

• Analyzing the topological structure of keyword graphs to predict potential links.
• Organizing groups of learners with negative correlation to promote sublearner diversity.
• Implementing a hierarchical negative correlation mechanism for layered subgraph feature extraction.

Main Results:

• The proposed model effectively identifies potential new research connections.
• The hierarchical negative correlation mechanism enhances pattern diversity across different research domains.
• The method demonstrates improved efficiency in terms of time and computational cost compared to existing ensemble techniques.

Conclusions:

• Link prediction, enhanced by diverse learner groups and hierarchical negative correlation, is a powerful tool for discovering new research ideas.
• This approach overcomes limitations of single learners in capturing diverse research patterns.
• The method offers a computationally efficient alternative for bibliometric analysis and innovation exploration.