A Recommender for Research Collaborators Using Graph Neural Networks.

Area of Science:

• Life Science and Biomedical Research
• Computational Science
• Data Science

Background:

• Effective researcher collaboration is crucial for scientific advancement.
• Identifying suitable collaborators is often challenging and time-consuming.
• A dedicated recommender system can streamline the process of finding research partners.

Purpose of the Study:

• To develop and evaluate a novel collaboration recommendation system for researchers.
• To leverage graph neural networks for capturing complex researcher relationships.
• To compare the performance of GNN-based recommenders against baseline methods.

Main Methods:

• Utilized the MEDLINE database for researcher and publication data.
• Developed a collaboration recommendation system using GraphSAGE and Temporal Graph Network (TGN).
• Implemented baseline models including LightGCN and gradient boosting trees for comparison.

Main Results:

• Graph neural network models demonstrated superior performance in collaboration recommendation.
• The proposed system effectively captures intrinsic, complex, and temporal dependencies among researchers.
• Both internal automatic evaluations and external end-user ratings confirmed the recommender's effectiveness.

Conclusions:

• Graph neural networks offer a powerful approach for building effective researcher collaboration recommendation systems.
• The developed system shows significant promise for enhancing collaboration in the life science and biomedical fields.
• Future work can explore further refinements and applications of GNNs in scientific collaboration.