A novel high applicability link prediction based on biased random walks

Complex systems are prevalent in nature, and link prediction is crucial for analyzing these systems. This method has gained attention for its ability to uncover various irregular movements with underlying similarities. While individual particle behavior is unpredictable, the collective behavior of large groups can be forecasted more accurately. The variability at lower scales vs universal similarities at higher scales is often overlooked. To address this, we investigate the random walk process on directed networks at a global scale. We propose an improved link prediction method using biased random walks to enhance accuracy and applicability. We first define out-degree neighbors as valid transmission options to prevent conflicts with in-degree paths. We then analyze how out-degrees and in-degrees affect particle transition probabilities, guiding particles toward high out-degrees and low in-degrees for a sustainable process. Finally, we use particle stabilization probabilities at different nodes as a similarity measure for predicting potential connections. Our method improves prediction precision and practicality, as validated by experiments on nine real-world networks, showing significant gains in accuracy and AUC scores.