Dual network embedding for representing research interests in the link prediction problem on co-authorship networks

Area of Science:

• Bibliometrics
• Network Science
• Information Retrieval

Background:

• Co-authorship networks offer insights into scientific collaboration.
• Effective collaborator recommendation requires understanding research interests and network structures.

Purpose of the Study:

• To develop a novel network representation for co-authorship networks using network embedding.
• To construct a recommender system for collaborator search based on shared research interests.
• To improve the prediction of future network structures.

Main Methods:

• Network embedding techniques applied to co-authorship data.
• Topic modeling for research paper analysis and keyword co-occurrence network construction.
• Link prediction (LP) model formulation for recommender system development.
• Integration of graph feature engineering and temporal network analysis.

Main Results:

• The proposed network representation model demonstrates superior performance in binary classification tasks on co-authorship networks.
• The system effectively generalizes author attributes by leveraging topic modeling and network embedding.
• The recommender system successfully identifies potential collaborators with similar research interests.

Conclusions:

• Network embedding combined with topic modeling provides a powerful approach for analyzing and understanding co-authorship networks.
• The developed link prediction model enhances the accuracy of collaborator recommendation systems.
• The study validates the model's effectiveness on a comprehensive 25-year dataset.