Systematic Prompt Design and Integration for LLM-Based Abstractive Summarization

Develop a systematic framework for designing prompts and integrating prompt engineering across the entire pipeline of large language model–based abstractive summarization to enable controllable abstraction levels and styles in generated summaries.

Background

LLMs have advanced abstractive summarization but often lack controllability over abstraction level and style, which limits alignment with diverse user needs and application contexts. Prompt engineering is a promising approach to guide models toward different levels of abstraction.

Despite this promise, the paper explicitly notes that establishing a systematic methodology for prompt design and its integration throughout the summarization process remains unresolved, motivating research into structured, end-to-end prompt frameworks that can reliably control summarization behavior.