Link Weight Prediction Using Weight Perturbation and Latent Factor.

Area of Science:

• Network science
• Machine learning
• Data mining

Background:

• Link weight prediction is crucial for network analysis, with applications in social networks, network modeling, and bioinformatics.
• Existing models often struggle to balance predictive performance with interpretability.
• The need for effective unsupervised methods in link weight prediction is significant.

Purpose of the Study:

• To develop an unsupervised mixed strategy for link weight prediction.
• To improve the balance between performance and interpretability in link weight prediction models.
• To leverage network structure and node attributes for accurate link weight estimation.

Main Methods:

• An unsupervised mixed strategy was employed for link weight prediction.
• The model utilizes the weighted adjacency matrix as direct input without preprocessing.
• The strategy combines network weight consistency with node-specific latent factors.

Main Results:

• The proposed scheme demonstrates competitive performance against state-of-the-art algorithms.
• Evaluation metrics include root-mean-square error and Pearson correlation coefficient.
• Extensive observations across numerous networks validate the method's efficacy.

Conclusions:

• Combining network weight consistency and latent node factors is highly effective for link weight prediction.
• The developed unsupervised strategy offers a promising approach for network analysis.
• This work contributes to advancing the field of network science and machine learning.