# A Discourse-Aware Attention Model for Abstractive Summarization of Long Documents

**Authors:** Arman Cohan, Franck Dernoncourt, Doo Soon Kim, Trung Bui, Seokhwan Kim, Walter Chang, Nazli Goharian  
**arXiv:** [1804.05685](https://arxiv.org/abs/1804.05685)  
**Date:** 2018-04-16  
**Summary page:** https://www.summarizepaper.com/en/paper/1804.05685v1/

## Key points

- Authors introduce a novel hierarchical encoder-decoder architecture for abstractive summarization of long documents
- Model incorporates discourse-aware attention mechanisms to focus on relevant parts of the document
- Empirical evaluation on scientific papers datasets shows superior performance in summary quality and effectiveness
- Model outperforms existing approaches, highlighting its potential for enhancing automatic summarization tasks
- Research contributes significantly to advancing abstractive summarization by addressing challenges of longer documents and showcasing benefits of incorporating discourse structure

## Summary

In their paper titled "A Discourse-Aware Attention Model for Abstractive Summarization of Long Documents," authors Arman Cohan, Franck Dernoncourt, Doo Soon Kim, Trung Bui, Seokhwan Kim, Walter Chang, and Nazli Goharian introduce a groundbreaking approach to abstractive summarization of lengthy documents such as research papers. Building upon the success of neural abstractive summarization models in handling shorter texts, the authors propose a novel hierarchical encoder-decoder architecture that takes into account the discourse structure of the input document. The key innovation in their model lies in the incorporation of discourse-aware attention mechanisms, which enable the decoder to focus on relevant parts of the document when generating the summary. This attention to discourse coherence results in more coherent and informative summaries compared to existing state-of-the-art models. The empirical evaluation conducted on two large-scale datasets of scientific papers demonstrates the superior performance of the proposed model. The results show a significant improvement in summary quality and overall effectiveness in capturing the essence of longer-form documents. The model's ability to outperform existing approaches highlights its potential for enhancing automatic summarization tasks in domains where detailed and comprehensive summaries are crucial. Overall, this research contributes significantly to advancing the field of abstractive summarization by addressing the unique challenges posed by longer documents and showcasing how incorporating discourse structure can lead to more accurate and contextually rich summaries.