Learning to Extract Coherent Summary via Deep Reinforcement Learning

TL;DR

This work addresses the challenge of producing coherent extractive summaries by introducing a neural coherence model and integrating it into a reinforcement-learning framework (RNES). The NES encoder captures long-document structure, while RNES optimizes a joint objective combining coherence and ROUGE-based informativeness. Empirical results on CNN/Daily Mail show state-of-the-art performance for ROUGE when balancing rewards, and human evaluations indicate improved coherence and readability. The approach demonstrates that coherence-aware reinforcement learning can meaningfully improve practical summary quality beyond traditional ROUGE optimization.

Abstract

Coherence plays a critical role in producing a high-quality summary from a document. In recent years, neural extractive summarization is becoming increasingly attractive. However, most of them ignore the coherence of summaries when extracting sentences. As an effort towards extracting coherent summaries, we propose a neural coherence model to capture the cross-sentence semantic and syntactic coherence patterns. The proposed neural coherence model obviates the need for feature engineering and can be trained in an end-to-end fashion using unlabeled data. Empirical results show that the proposed neural coherence model can efficiently capture the cross-sentence coherence patterns. Using the combined output of the neural coherence model and ROUGE package as the reward, we design a reinforcement learning method to train a proposed neural extractive summarizer which is named Reinforced Neural Extractive Summarization (RNES) model. The RNES model learns to optimize coherence and informative importance of the summary simultaneously. Experimental results show that the proposed RNES outperforms existing baselines and achieves state-of-the-art performance in term of ROUGE on CNN/Daily Mail dataset. The qualitative evaluation indicates that summaries produced by RNES are more coherent and readable.

Figures (1)

• Figure 1: Illustration of neural coherence model which is built upon ARC-II proposed by NIPS2014_hu.