基于时序网络节点嵌入的影响力最大化算法

A node-embedding-based influence maximization algorithm in temporal network

随着互联网的高速发展,影响力最大化算法在病毒营销、谣言遏制等众多领域得到广泛应用,它旨在搜索一个指定规模的种子节点集,以实现影响力传播范围最大化.尽管已有许多影响力最大化算法被提出,复杂系统的时序特征却鲜有被纳入考量.为了解决时序网络中的影响力最大化问题,本文提出了一种基于时序网络节点嵌入的影响力最大化算法(Micro-and Macro-based Clustering,MMC).该算法能够通过节点嵌入有效地学习时序网络中的微观和宏观维度上的动力学特征.基于嵌入向量,对时序网络进行簇聚类,以降低影响力最大化算法的复杂度,减小种子节点集间的影响力范围重叠,从而进一步提升算法的效果.为了验证MMC的有效性,本文在真实时序网络中进行了影响力传播实验与相似性比较.结果显示,相较于其他5种对比算法,MMC筛选出的种子节点在保证较好的传播效果的同时,能够跳出既有算法的框架.参数敏感性的分析表明,节点聚类过程有效地提升了算法效果.

The rapid growth of social networks has resulted in the increased application of influence maximization algorithms in various fields,such as viral marketing and rumor containment.It searches for a set of seed nodes of a given size that can maximize the range of influence propagation.Several influence maximization algorithms have been proposed;however,the temporal characteristics of complex systems are rarely considered.To address this issue,we propose an influence maximization algorithm based on temporal network node embedding called micro-and macro-based clustering(MMC).The proposed algorithm can effectively learn the micro-and macro-dynamic characteristics of temporal networks by node embedding.We cluster the temporal network based on the embedding vectors to decrease the complexity of the influence maximization algorithm and the overlapping of influence spread among seed node sets,thereby further improving the performance of the algorithm.Furthermore,we conducted diffusion and similarity comparison experiments on real-world temporal networks to validate the effectiveness of MMCs.The results revealed that compared to five other comparative algorithms,the seed nodes selected by MMC not only guarantee greater influence spread but also go beyond the framework of existing algorithms.The analysis of parameter sensitivity revealed that effective clustering of nodes improves algorithm performance.