Improving visual-semantic embeddings by learning semantically-enhanced hard negatives for cross-modal information retrieval - Citation Graph | Papersgraph