Co-Regularized PLSA for Multi-Modal Learning