Graph-based Neural Sentence Ordering
Abstract
Sentence ordering is to restore the original paragraph from a set of sentences. It involves capturing global dependencies among sentences regardless of their input order. In this paper, we propose a novel and flexible graph-based neural sentence ordering model, which adopts graph recurrent network \cite{Zhang:acl18} to accurately learn semantic representations of the sentences. Instead of assuming connections between all pairs of input sentences, we use entities that are shared among multiple sentences to make more expressive graph representations with less noise. Experimental results show that our proposed model outperforms the existing state-of-the-art systems on several benchmark datasets, demonstrating the effectiveness of our model. We also conduct a thorough analysis on how entities help the performance.
Cite
@article{arxiv.1912.07225,
title = {Graph-based Neural Sentence Ordering},
author = {Yongjing Yin and Linfeng Song and Jinsong Su and Jiali Zeng and Chulun Zhou and Jiebo Luo},
journal= {arXiv preprint arXiv:1912.07225},
year = {2019}
}
Comments
IJCAI2019