Document clustering using word clusters via the information bottleneck method