Analyzing and Securing Social Networks

199Social Network Classification through Data Partitioning

We see slightly different results for experiments with the 20% replication rate. In the 50%

experiment, we again see that domain specic is signicantly more accurate than either of the

other metrics. However, what we can see from each graph is that the largest jump comes from the

change from a single partition to two. We see that there is a signicant difference at even the two-

partitions entry for the same partitioning metric. This should be attributable only to the greater

replication rate, since that is the only difference in the two experiments. One would expect that

differences attributable to the random selection of the training set would be less signicant over

the multiple runs ...