Multi-Modal Adversarial Autoencoders for Recommendations of Citations and Subject Labels