Fig. 1. (Color online) The blog post corpus was segmented for model evaluation into an open set with no shared ID labels and a closed set with shared labels. Both sets were evaluated using the same evaluation set comprised of users in the closed set.
© NPSM