基于初始信息流和层间相关性的多层网络链路预测

Link Prediction in Multiplex Networks Based on Initial Information Flow and Inter-Layer Relevance

摘要: 链路预测是理解多层网络演化机制的核心任务。针对多层网络链路预测中忽略层间拓扑差异与节点影响力异质性的问题，本文提出基于改进多层初始信息流(M-IIF)与层间相关性的综合预测框架。首先，引入阻尼因子η改进IIF指标，通过模拟能量耗散精准度量节点传播潜能，有效抑制“超级节点”导致的数值畸变；其次，利用层间结构相似度构建自适应加权机制，实现多源信息的鲁棒融合；最后，通过线性解耦模型整合层内局部与半全局特征，计算最终相似性得分。多组真实数据集实验验证了该方法在Precision指标上的优越性。研究表明，本文算法在保持高精度的同时，具备较低的时间复杂度与良好的可扩展性，适用于大规模多层网络的链路预测任务。

Abstract: Link prediction is a fundamental task for understanding the evolutionary mechanisms of multilayer networks. Addressing the issues of overlooking interlayer topological differences and node influence heterogeneity in multilayer link prediction, this paper proposes a comprehensive prediction framework based on Modified Multilayer Initial Information Flow (M-IIF) and interlayer relevance. First, a damping factor η is introduced to refine the IIF indicator; by simulating the physical process of energy dissipation, the node spreading potential is measured with high precision, which effectively suppresses numerical distortions induced by “super-nodes”. Second, an adaptive weighting mechanism is constructed based on interlayer structural similarity to achieve the robust fusion of multi-source information. Finally, a linear decoupling model is developed to integrate intra-layer local and semi-global features for the calculation of final similarity scores. Experimental results on multiple real-world datasets demonstrate the superiority of the proposed method, particularly in terms of the Precision metric. The study indicates that the proposed algorithm maintains high predictive accuracy while possessing low computational complexity and favorable scalability, rendering it highly suitable for link prediction tasks in large-scale multilayer networks.