Link prediction benchmark datasets

Question

I'd like to test a new algorithm for link prediction on graphs. I'm interested in both weighted/unweighted and directed/undirected graphs.

What are some common benchmark datasets that researchers often use to test their algorithms? I know that within Computer Vision people often use MNIST or CIFAR, but I haven't found similar datasets for link prediction.

Answer 1

A cursory Google search gave me:

• http://noesis.ikor.org/datasets/link-prediction
• http://ial.eecs.ucf.edu/travian.php
• http://konect.uni-koblenz.de/networks/

Another way would be to create your own dataset by downloading say Github user's graph, defining a network by adding edges between user A and user B if A follows B and so on. Then you can take snapshots of this graph at different time instances and see if you can use last t snapshots to predict the edges in the t+1th snapshot. Of course that would be a lot more extra effort and the links I posted may suffice. Please let me know if it helps.