添加 Link prediction metric

[unccx]

Link prediction也是分类任务，模仿节点分类任务的metric添加accuracy，f1_score，confusion_matrix，auc。其中accuracy和 F1 分数的截止阈值选择所有测试超链接分数的平均值，这一点是参考论文 A Survey on Hyperlink Prediction 中提到的：

We first considered a negative sampling strategy with α = 0.5 and β = 1 as used in NHP. For each network, we randomly split the hyperlink set including positive and negative hyperlinks into training and testing sets by a ratio of 3:2 over 10 trials (no negative sample was introduced in the training set for HPRA and C3MM). We trained the six learning models on the training set and tested on the testing set. The results are shown in Table IV, where each value is the mean over 10 trials, and we picked the mean of all the testing hyperlink scores as the cutoff threshold for computing the F1 score.