Generative Adversarial Network for Abstractive Text Summarization

TL;DR

This paper introduces an adversarial training framework for abstractive text summarization, combining reinforcement learning-based generation with a discriminator to improve summary quality and diversity.

Contribution

It presents a novel adversarial approach that jointly trains a generator and discriminator for abstractive summarization, enhancing the abstraction and diversity of generated summaries.

Findings

Achieves competitive ROUGE scores on CNN/Daily Mail dataset

Generates more abstractive and diverse summaries

Produces more readable summaries

Abstract

In this paper, we propose an adversarial process for abstractive text summarization, in which we simultaneously train a generative model G and a discriminative model D. In particular, we build the generator G as an agent of reinforcement learning, which takes the raw text as input and predicts the abstractive summarization. We also build a discriminator which attempts to distinguish the generated summary from the ground truth summary. Extensive experiments demonstrate that our model achieves competitive ROUGE scores with the state-of-the-art methods on CNN/Daily Mail dataset. Qualitatively, we show that our model is able to generate more abstractive, readable and diverse summaries.