On Extractive and Abstractive Neural Document Summarization with Transformer Language Models

TL;DR

This paper introduces a neural summarization method combining extractive and abstractive techniques using transformer models, significantly improving summary quality for long documents.

Contribution

It proposes a simple extractive step to enhance transformer-based abstractive summarization, leading to more abstractive summaries with higher ROUGE scores.

Findings

Extractive step improves summarization results

Method produces more abstractive summaries than prior work

Achieves higher ROUGE scores

Abstract

We present a method to produce abstractive summaries of long documents that exceed several thousand words via neural abstractive summarization. We perform a simple extractive step before generating a summary, which is then used to condition the transformer language model on relevant information before being tasked with generating a summary. We show that this extractive step significantly improves summarization results. We also show that this approach produces more abstractive summaries compared to prior work that employs a copy mechanism while still achieving higher rouge scores. Note: The abstract above was not written by the authors, it was generated by one of the models presented in this paper.