Integrating word embeddings and document topics with deep learning in a video classification framework.