Related Experiment Videos
SAF: A Spectral-Adaptive Fusion Algorithm for Link Prediction in Complex Networks
Wen Liang1,2, Chunyu Yang1, Qiwei Liu3
1College of Information Science and Engineering, Shenyang Ligong University, Shenyang 110159, China.
Entropy (Basel, Switzerland)
|July 28, 2026
Summary
This study introduces the spectral-adaptive fusion (SAF) algorithm for network link prediction. SAF effectively combines global and local network features, improving accuracy and reducing runtime compared to existing methods.
Area of Science:
- Network Science
- Data Mining
- Machine Learning
Background:
- Link prediction is vital for understanding complex networks and applications like social recommendation.
- Existing methods struggle to effectively integrate both global and local network structural information.
Purpose of the Study:
- To propose a novel algorithm, spectral-adaptive fusion (SAF), for accurate link prediction.
- To enhance the exploitation of both global and local network structural information.
Main Methods:
- SAF constructs a spectral embedding matrix by selecting key spectral components.
- It derives row-column normalized and Gaussian kernel matrices, adaptively fused using a common-neighbor mechanism.
- A truncated ratio of 5% was determined via energy retention and spectral gap analyses.
Main Results:
- SAF achieved an average runtime reduction of 71.0% across eight datasets.
- It showed a 2.22% improvement over graph neural networks and 10.65% over matrix factorization methods in AUC.
- SAF maintained AUPR values above 0.91 on four networks even at low training ratios.
Conclusions:
- The spectral-adaptive fusion (SAF) algorithm demonstrates robust and effective link prediction capabilities.
- SAF successfully balances global and local network features while mitigating the impact of central nodes.
- The method offers significant improvements in accuracy and efficiency for network link prediction tasks.
