Improving Textual Network Embedding with Global Attention via Optimal Transport

39450

Knowledge Triples