Fact-level Extractive Summarization with Hierarchical Graph Mask on BERT

TL;DR

This paper introduces a fact-level extractive summarization method using a hierarchical graph mask on BERT, improving the alignment with human summaries and achieving state-of-the-art results on CNN/DailyMail.

Contribution

It proposes a novel fact-level extraction approach with a hierarchical structure and a graph mask integrated into BERT, enhancing extractive summarization performance.

Findings

Achieves state-of-the-art results on CNN/DailyMail dataset

Effectively models multi-level textual granularities

Improves alignment with human-written summaries

Abstract

Most current extractive summarization models generate summaries by selecting salient sentences. However, one of the problems with sentence-level extractive summarization is that there exists a gap between the human-written gold summary and the oracle sentence labels. In this paper, we propose to extract fact-level semantic units for better extractive summarization. We also introduce a hierarchical structure, which incorporates the multi-level of granularities of the textual information into the model. In addition, we incorporate our model with BERT using a hierarchical graph mask. This allows us to combine BERT's ability in natural language understanding and the structural information without increasing the scale of the model. Experiments on the CNN/DaliyMail dataset show that our model achieves state-of-the-art results.