一种融合局部拓扑影响力的时序链路预测算法

链路预测旨在发现复杂网络中的未知连接和未来可能的连接,在推荐系统等实际应用中具有重要作用.考虑到许多真实网络的时序特性,时序链路预测逐渐成为研究热点.当前,基于时间序列分析的方法往往忽略了网络演化过程对网络本身的影响,而基于静态网络演化的方法大多仅考虑了局部连边的演化影响,对网络拓扑结构的演化特性挖掘有限.针对上述问题...

A Temporal Link Predict Algorithm Based on Fusion Local Structure Influence

Link prediction aims to discover missing connected edges and possible future interaction in complex networks. The evolution mechanism of temporal networks has gained the attention of researchers with its ubiquitous applications in a variety of real-world scenarios. At present, many methods based on time series analysis are proposed, but the influence of the network evolution process on the network itself is ignored, and the methods based on the static network algorithm only consider the influence of the evolution of edges, which may lead to inadequate utilization of feature information and can not achieve better prediction accuracy. In view of the above problems, a novel Temporal Link Prediction algorithm base on Fusion Local Structure Influence (TLP-FLSI) is proposed, which fuses the impact of local nodes and edges. Firstly, based on the influence of network topology structure, Common Temporal Link Prediction Model(CTLPM)is proposed. Secondly, the evolution mechanism of the interaction between topological entities on the dynamic network is studied, and the evolution factors of nodes and edges, as well as the decay evolution factors of time series are defined respectively, and considering various factors, TLP-FLSI is derivated from CTLPM. Finally, compared with traditional temporal link predict method, including moving average methods, error correction methods, extended weighted method, graph attention methods, experimental results of seven real data sets show that TLP-FLSI achieves great improvement in accuracy and ranking score.