基于深度强化学习的复杂网络节点影响力排序算法

A Node Influence Ranking Algorithm in Complex Networks Based on Deep Reinforcement Learning

针对当前节点影响力评估算法准确度较低的情况,提出了一种基于深度强化学习的节点影响力排序算法。该算法从网络拆解的视角看待节点影响力,将节点影响力的排序问题转换为网络拆除策略的优化问题。算法首先利用排序学习训练图神经网络模型的节点特征提取能力,然后使用强化学习对依赖于网络状态的节点断连行为做价值学习,最后使用训练完成的模型预测网络拆除的最佳策略,即节点影响力的最准确排序。仿真实验证明,所提算法在典型真实数据集的CN(Crtical Node)与ND(Network Dismantling)问题上,相较于PageRank算法,准确度分别提升了31.1%与29.0%。同时,该算法具有较低的复杂度,可为网络稳定性分析和网络性能优化提供技术支撑。

Aiming at the low accuracy of current node influence assessment algorithms,a node influence ranking algorithm based on deep reinforcement learning is proposed.The algorithm views node influence from the perspective of network dismantling,and converts the problem of ranking node influence into the optimization problem of network dismantling strategy.This algorithm first uses Learning-To-Rank to train the node feature extraction ability of graph neural network model,then uses reinforcement learning to do value learning on node disconnection actions that depend on the network state,and finally uses the trained model to predict the optimal strategy for network dismantling,that is the most accurate ordering of node influence.Simulation experiments prove that the proposed algorithm improves the accuracy by 31.1%and 29.0%compared with PageRank algorithm on critical nod(CN)and network dismantling(ND)problems with typical real datasets,respectively.Meanwhile,the algorithm has low complexity and can provide technical support for network stability analysis and network performance optimization.