Augmented Abstractive Summarization With Document-LevelSemantic Graph

TL;DR

This paper introduces a semantic graph-enhanced abstractive summarization method that extracts key entities, constructs a graph, and uses a combined Bi-LSTM and graph encoder to improve summary quality, validated by evaluations.

Contribution

It proposes a novel approach integrating semantic graphs with neural decoding for improved abstractive summarization performance.

Findings

Enhanced summary quality demonstrated by automatic metrics

Human evaluations confirm improved relevance and coherence

Effective use of entity graphs in summarization tasks

Abstract

Previous abstractive methods apply sequence-to-sequence structures to generate summary without a module to assist the system to detect vital mentions and relationships within a document. To address this problem, we utilize semantic graph to boost the generation performance. Firstly, we extract important entities from each document and then establish a graph inspired by the idea of distant supervision \citep{mintz-etal-2009-distant}. Then, we combine a Bi-LSTM with a graph encoder to obtain the representation of each graph node. A novel neural decoder is presented to leverage the information of such entity graphs. Automatic and human evaluations show the effectiveness of our technique.