BERT4SO: Neural Sentence Ordering by Fine-tuning BERT

TL;DR

BERT4SO introduces a novel approach to sentence ordering by fine-tuning BERT with special token representations and a listwise ranking loss, achieving improved results on benchmark datasets.

Contribution

The paper presents a new BERT-based method for sentence ordering that enhances sentence interactions and optimizes with a margin-based listwise ranking loss.

Findings

Effective on five benchmark datasets

Outperforms previous neural ranking models

Demonstrates the benefit of segment embeddings and special tokens

Abstract

Sentence ordering aims to arrange the sentences of a given text in the correct order. Recent work frames it as a ranking problem and applies deep neural networks to it. In this work, we propose a new method, named BERT4SO, by fine-tuning BERT for sentence ordering. We concatenate all sentences and compute their representations by using multiple special tokens and carefully designed segment (interval) embeddings. The tokens across multiple sentences can attend to each other which greatly enhances their interactions. We also propose a margin-based listwise ranking loss based on ListMLE to facilitate the optimization process. Experimental results on five benchmark datasets demonstrate the effectiveness of our proposed method.