# Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization

**Authors:** Shashi Narayan, Shay B. Cohen, Mirella Lapata  
**arXiv:** [1808.08745](https://arxiv.org/abs/1808.08745)  
**Date:** 2018-08-27  
**Summary page:** https://www.summarizepaper.com/en/paper/1808.08745v1/

## Key points

- Extreme summarization is a new single-document summarization task
- It requires an abstractive modeling approach
- The goal is to generate a concise one-sentence summary that answers the question "What is the article about?"
- The authors gathered a large-scale dataset of online articles from the BBC
- They propose a novel abstractive model based on convolutional neural networks and conditioned on the article's topics
- Their architecture effectively captures long-range dependencies in documents and identifies relevant content
- Their model outperforms both an oracle extractive system and state-of-the-art abstractive approaches when evaluated by automated metrics and human judges

## Summary

In their paper titled "Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization," authors Shashi Narayan, Shay B. Cohen, and Mirella Lapata introduce extreme summarization as a new single-document summarization task that requires an abstractive modeling approach rather than extractive strategies. The goal is to generate a concise one-sentence summary that answers the question "What is the article about?". To facilitate this task, they gather a large-scale dataset of online articles from the British Broadcasting Corporation (BBC). The authors propose a novel abstractive model based on convolutional neural networks and conditioned on the article's topics. Through experiments, they demonstrate that their architecture effectively captures long-range dependencies in documents and identifies relevant content. Their model outperforms both an oracle extractive system and state-of-the-art abstractive approaches when evaluated by both automated metrics and human judges.