Improving word mover's distance by leveraging self-attention matrix

Open in new window