基于预测误差修正的时序链路预测方法

论文主要针对时序链路预测方法进行研究。分析了静态链路预测方法的弊端,认为忽视网络演化趋势信息会对链路预测产生负面影响;还提出了链路预测误差的概念用于描述网络趋势信息,并以此为基础提出一种基于预测误差修正的时序链路预测方法。该方法首先对待预测时刻之前一个时间窗口内的多幅网络图分别采用静态链路预测方法进行预测,记录每次的预测误差并计算其修正值,最后对待测时刻静态预测结果进行修正得到最终预测结果。通过在两个真实网络数据集上进行系列实验表明,该方法较大提升了静态链路预测方法的预测精确度,与另一种典型的时序链路预测方法相比其精度也有所提升,且算法时间复杂度较低。另外,实验中还发现链路预测误差序列与网络链路总数序列存在"镜面对称"关系,分析其内在原因证明了所提方法的普适性。

A Temporal Link Prediction Method Based on Link Prediction Error Correction

The temproral link prediction method is investigated in this paper. The disadvantages of the static link prediction methods are analyzed, considering that ignoring the evolving information of networks will lead to a negative impact on link predicting. The concept of link prediction error is proposed to describe the evolving information of networks, and a temporal link prediction method is proposed based on the prediction error correction. Firstly, several static link prediction are carried out using each graph in the previous periods window, and then the prediction errors are recorded and used for calculating the modification value. At last, the final prediction result is acquired through refining the static prediction result with the modification value. Several experiments are conducted using two real network datasets. The results show that the proposed method achieves better performance than the static link prediction methods and a typical temporal link prediction method. In addition, it can be found that a relation of ‘mirror symmetry’ exists between prediction error series and total link number series, which demonstrates the universality of the proposed method.