ICE: Information coverage estimate for automatic evaluation abstractive summaries

作者：

Highlights：

• ICE metric assesses the extent to which a summary retains critical information.

• ICE is a similarity-based measure for rating abstractive summaries.

• It incorporates relative-closeness (synonymy) between words.

• ICE implements POS-based Keyword Sampling for efficient estimation.

摘要

•ICE metric assesses the extent to which a summary retains critical information.•ICE is a similarity-based measure for rating abstractive summaries.•It incorporates relative-closeness (synonymy) between words.•ICE implements POS-based Keyword Sampling for efficient estimation.

论文关键词：Abstractive text summarization (ATS),Cosine-similarity,Keyword extraction,POS tagging,Word embeddings

论文评审过程：Received 6 March 2021, Revised 22 July 2021, Accepted 9 October 2021, Available online 30 October 2021, Version of Record 3 November 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2021.116064