On Semi-Supervised Multiple Representation Behavior Learning