Salient Information Prompting to Steer Content in Prompt-based Abstractive Summarization

TL;DR

This paper demonstrates that incorporating phrase-level salient keyphrases into prompts significantly improves the quality of LLM-generated summaries, with a new lightweight keyphrase extractor enhancing performance across models and datasets.

Contribution

It introduces SigExt, a lightweight, finetuneable keyphrase extractor, and shows that salient phrase-level information in prompts enhances summarization quality in prompt-based LLMs.

Findings

Adding keyphrases improves ROUGE scores and recall.

Phrase-level salient info outperforms word- or sentence-level info.

Salient information can be effectively extracted with SigExt across models and datasets.

Abstract

Large language models (LLMs) can generate fluent summaries across domains using prompting techniques, reducing the need to train models for summarization applications. However, crafting effective prompts that guide LLMs to generate summaries with the appropriate level of detail and writing style remains a challenge. In this paper, we explore the use of salient information extracted from the source document to enhance summarization prompts. We show that adding keyphrases in prompts can improve ROUGE F1 and recall, making the generated summaries more similar to the reference and more complete. The number of keyphrases can control the precision-recall trade-off. Furthermore, our analysis reveals that incorporating phrase-level salient information is superior to word- or sentence-level. However, the impact on hallucination is not universally positive across LLMs. To conduct this analysis, we introduce Keyphrase Signal Extractor (SigExt), a lightweight model that can be finetuned to extract salient keyphrases. By using SigExt, we achieve consistent ROUGE improvements across datasets and open-weight and proprietary LLMs without any LLM customization. Our findings provide insights into leveraging salient information in building prompt-based summarization systems. We release our code at \url{https://github.com/amazon-science/SigExt}