Neighbor-Enhanced Link Prediction in Bipartite Networks

Area of Science:

• Network Science
• Data Mining
• Computational Social Science

Background:

• Link prediction in bipartite networks is complex due to unique structural constraints.
• Existing methods often struggle with heterogeneous node degree distributions, leading to prediction bias.
• Local structural methods are popular but can be undermined by degree heterogeneity.

Purpose of the Study:

• To develop a novel link prediction framework for bipartite networks.
• To address and mitigate the bias caused by heterogeneous node degrees.
• To improve the accuracy and robustness of link prediction in complex networks.

Main Methods:

• Proposed a framework that adjusts for degree heterogeneity of intermediate nodes.
• Incorporated influence of intermediate nodes within local connection patterns.
• Utilized quadrangle graphs to differentiate node roles and capture network properties.

Main Results:

• The framework effectively mitigates degree bias in bipartite networks.
• Demonstrated considerable improvements in link prediction accuracy.
• Achieved competitive and robust performance across ten diverse bipartite networks.

Conclusions:

• The novel framework offers a significant advancement in bipartite network link prediction.
• Explicitly addressing degree heterogeneity is crucial for accurate link prediction.
• The method shows strong potential for real-world applications involving complex network analysis.