基于多种信息组合模式的非负矩阵分解链路预测模型

针对基于拓扑结构相似性的链路预测算法中网络稀疏性和噪声问题,提出一种基于非负矩阵分解的链路预测模型。该模型从微观与宏观两个层面出发,融合网络内部和外部的辅助信息,减轻了网络稀疏性造成的影响,提升了算法预测的整体性能。提出的三种信息组合模式体现出宏观与微观角度下的信息融合策略。在13个真实网络数据集上的实验结果展示了算法预测性能上的优越性。

Based on nonnegative matrix factorization for link prediction model via multiple information combination modes

Aiming at network sparsity and noise problem in link prediction algorithm based on topological structure similarity,this paper proposed a link prediction model based on non-negative matrix decomposition.The model started from the micro and macro levels,and simultaneously integrated the internal and external auxiliary information of the network,which mitigated the impact of network sparsity and improved the overall performance of the algorithm prediction.The proposed three information combination models reflected the information fusion strategy from macro and micro perspectives.The experimental results on 13 real network datasets demonstrate the superiority of algorithm prediction performance.