基于链路预测算法分析虚假链接问题

链路预测与虚假链接是复杂网络的两大研究热点,目前为止,链路预测方法的研究已经非常成熟,而对于虚假链接的研究却仍旧没有得到太多的关注.根据链路预测与虚假链接的联系,用17种经典的链接预测算法,研究了这些算法在8个真实网络数据集中的识别虚假链接效果.实验结果表明,这些算法在识别虚假链接中的AUC值明显低于在链接预测中的AUC值;并且随着测试集比例的增加,在给定某个算法和数据集上,AUC值逐渐降低.

An analysis of the spurious link based on the link prediction algorithm

Link prediction and spurious links are two research hotspots of the complex network. So far,the research of the link prediction method has been regarded as very mature,but the research on spurious links still has not received much attention. Based on the correlation between link prediction and spurious links,this paper uses 17 classical link prediction algorithms,which reveal their validity in identifying the spurious link in eight real network datasets. The results of the experiment show that the values of AUC are much smaller in the spurious link than those in the link prediction; with the ratio of the test sets increasing,the values of AUC will decrease in the given algorithm and the real data set.