Skip to content

Latest commit

 

History

History
14 lines (8 loc) · 690 Bytes

README.md

File metadata and controls

14 lines (8 loc) · 690 Bytes

A Cascaded Approach to Neural Abstractive Summarization with Content Selection and Fusion

Data for the AACL-IJCNLP 2020 paper "A Cascaded Approach to Neural Abstractive Summarization with Content Selection and Fusion"

Data

Our data consists of > 1 million sentence fusion instances, of the form:

Input: one or two articles sentences + token-level highlights indicating which tokens were used to create the output summary sentence

Output: the summary sentence formed by compressing/fusing the input sentences

Our data is derived from the CNN/Daily Mail summarization dataset.

Link to our data: https://www.dropbox.com/sh/227japgc3q3klkd/AABwwFi6LueV8F6sokjHqSsTa?dl=0