Our ICML2012 and CIKM2013 papers used the same basic data, although the CIKM2013 paper used more trials. The data below includes nodes, their attributes, links, and true labels for each node. Also, information on the actual splits used in our CIKM2013 paper are included.
Most of our code is in Java and fairly intertwined with the Proximity toolkit and a large amount of instrumentation, which makes it hard to usefully share. However, code to perform hybrid label regularization (from the ICML2012 paper) is available upon request; it is an extension to Weka's logistic regression module. I'm also very willing to answer any questions you might have about how to reproduce our methods.