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

We introduce extreme summarization, a new single-document summarization task which does not favor extractive strategies and calls for an abstractive modeling approach. The idea is to create a short, one-sentence news summary answering the question "What is the article about?". We collect a real-world, large-scale dataset for this task by harvesting online articles from the British Broadcasting Corporation (BBC). We propose a novel abstractive model which is conditioned on the article's topics and based entirely on convolutional neural networks. We demonstrate experimentally that this architecture captures long-range dependencies in a document and recognizes pertinent content, outperforming an oracle extractive system and state-of-the-art abstractive approaches when evaluated automatically and by humans.

PDF Abstract EMNLP 2018 PDF EMNLP 2018 Abstract

Datasets


Introduced in the Paper:

XSum

Used in the Paper:

NEWSROOM

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Text Summarization X-Sum T-ConvS2S ROUGE-1 31.89 # 9
ROUGE-2 11.54 # 12
ROUGE-3 25.75 # 5
Text Summarization X-Sum Baseline : Random ROUGE-1 15.16 # 15
ROUGE-2 1.78 # 17
ROUGE-3 11.27 # 11
Text Summarization X-Sum Baseline : Lead-3 ROUGE-1 16.30 # 14
ROUGE-2 1.60 # 18
ROUGE-3 11.95 # 10
Text Summarization X-Sum Baseline : Extractive Oracle ROUGE-1 29.79 # 10
ROUGE-2 8.81 # 14
ROUGE-3 22.66 # 7
Text Summarization X-Sum PtGen-Covg ROUGE-1 28.10 # 13
ROUGE-2 8.02 # 16
ROUGE-3 21.72 # 9
Text Summarization X-Sum Seq2Seq ROUGE-1 28.42 # 12
ROUGE-2 8.77 # 15
ROUGE-3 22.48 # 8
Text Summarization X-Sum PtGen ROUGE-1 29.70 # 11
ROUGE-2 9.21 # 13
ROUGE-3 23.24 # 6

Methods


No methods listed for this paper. Add relevant methods here