TriSum: Learning Summarization Ability from Large Language Models with Structured Rationale
This work addresses the challenge of using LLMs for text summarization in scenarios with limited computational resources and privacy concerns, offering an incremental improvement over existing distillation methods.
The paper tackles the problem of deploying large language models for text summarization in resource-constrained and privacy-centric settings by introducing TriSum, a framework that distills LLMs' abilities into a compact local model, achieving performance gains of 4.5%, 8.5%, and 7.4% on benchmarks like CNN/DailyMail, XSum, and ClinicalTrial.
The advent of large language models (LLMs) has significantly advanced natural language processing tasks like text summarization. However, their large size and computational demands, coupled with privacy concerns in data transmission, limit their use in resource-constrained and privacy-centric settings. To overcome this, we introduce TriSum, a framework for distilling LLMs' text summarization abilities into a compact, local model. Initially, LLMs extract a set of aspect-triple rationales and summaries, which are refined using a dual-scoring method for quality. Next, a smaller local model is trained with these tasks, employing a curriculum learning strategy that evolves from simple to complex tasks. Our method enhances local model performance on various benchmarks (CNN/DailyMail, XSum, and ClinicalTrial), outperforming baselines by 4.5%, 8.5%, and 7.4%, respectively. It also improves interpretability by providing insights into the summarization rationale.