Many real world, complex phenomena have underlying structures of evolvingnetworks where nodes and links are added and removed over time. A centralscientific challenge is the description and explanation of network dynamics,with a key test being the prediction of short and long term changes. For theproblem of short-term link prediction, existing methods attempt to determineneighborhood metrics that correlate with the appearance of a link in the nextobservation period. Recent work has suggested that the incorporation oftopological features and node attributes can improve link prediction. Weprovide an approach to predicting future links by applying the CovarianceMatrix Adaptation Evolution Strategy (CMA-ES) to optimize weights which areused in a linear combination of sixteen neighborhood and node similarityindices. We examine a large dynamic social network with over $10^6$ nodes(Twitter reciprocal reply networks), both as a test of our general method andas a problem of scientific interest in itself. Our method exhibits fastconvergence and high levels of precision for the top twenty predicted links.Based on our findings, we suggest possible factors which may be driving theevolution of Twitter reciprocal reply networks.
展开▼
