Collaborative Recurrent Autoencoder: Recommend while Learning to Fill in the Blanks