Robust Neural Abstractive Summarization Systems and Evaluation against Adversarial Information

TL;DR

This paper introduces a semantic-aware neural abstractive summarization model that improves factual accuracy and robustness against adversarial content, outperforming existing models in identifying off-topic information and producing more faithful summaries.

Contribution

The paper presents a novel semantic interpretation approach for neural summarization and an adversarial evaluation scheme to enhance factual correctness and robustness.

Findings

Our model outperforms pointer-generator in adversarial tests.

Human evaluation shows more informative and faithful summaries.

The approach reduces redundancy in generated summaries.

Abstract

Sequence-to-sequence (seq2seq) neural models have been actively investigated for abstractive summarization. Nevertheless, existing neural abstractive systems frequently generate factually incorrect summaries and are vulnerable to adversarial information, suggesting a crucial lack of semantic understanding. In this paper, we propose a novel semantic-aware neural abstractive summarization model that learns to generate high quality summaries through semantic interpretation over salient content. A novel evaluation scheme with adversarial samples is introduced to measure how well a model identifies off-topic information, where our model yields significantly better performance than the popular pointer-generator summarizer. Human evaluation also confirms that our system summaries are uniformly more informative and faithful as well as less redundant than the seq2seq model.