Neural Sentence Ordering Based on Constraint Graphs

Sentence ordering aims at arranging a list of sentences in the correct order. Based on observation that sentence order different distances may rely types information, we devise new approach based multi-granular orders between sentences. These form multiple constraint graphs, which are then encoded by Graph Isomorphism Networks and fused into representations. Finally, is determined using order-enhanced Our experiments five benchmark datasets show our method outperforms all existing baselines significantly, achieving state-of-the-art performance. The results demonstrate advantage considering information graph neural networks to integrate content for task. code available https://github.com/DaoD/ConstraintGraph4NSO.