Skip to main content
Fig. 3 | Computational Social Networks

Fig. 3

From: Coevolution of a multilayer node-aligned network whose layers represent different social relations

Fig. 3

We plot the ROC curves for prediction based on different thresholds of number of calls, messages, and collocations. With a threshold of zero, the false positive rate is 100% and true positive rate (recall) is 100% as well. Moreover, with the threshold equal to the maximum value, the false positive rate is 0% and true positive rate (recall) is 0% as well. We observe that the predictability is significantly higher than random. We see that there is a sharp drop in the false positive rate (which means, increase in accuracy) accompanied by a sharp drop in the recall when the thresholds increase above a certain limit. Behavioral edges where the number of calls, messages, or collocations is above a certain threshold are classified as significant contacts. We observe that as the threshold increases, the false positive rate (FPR) and the true positive rate (TPR) decrease gradually. At a certain higher value of the threshold, the rate of change of both the FPR and TPR increases significantly. This reflects on the distribution of the values of behavioral weights (be it the number of calls, messages, or collocations) of nominees vs. non-nominees and to-be-nominees vs. not-to-be-nominees. Nominees and to-be-nominees are much more likely to have higher behavioral weights, non-nominees and not-to-be-nominees can have higher behavioral weights, but less often. Also, nominees and to-be-nominees are much less likely to have very low behavioral weights

Back to article page