Enhancing Incremental Summarization with Structured Representations

TL;DR

This paper introduces structured knowledge representations and a dynamic update strategy to improve incremental summarization in large language models, achieving significant performance gains over previous unstructured methods.

Contribution

The paper presents a novel structured memory format ($GU_{json}$) and the Chain-of-Key ($CoK_{json}$) strategy for dynamic updates, advancing incremental summarization techniques.

Findings

Improved summarization performance by 40% and 14% on two datasets.

Dynamic updates with CoK_{json} further boost performance by 7% and 4%.

Structured representations reduce information overload in LLMs.

Abstract

Large language models (LLMs) often struggle with processing extensive input contexts, which can lead to redundant, inaccurate, or incoherent summaries. Recent methods have used unstructured memory to incrementally process these contexts, but they still suffer from information overload due to the volume of unstructured data handled. In our study, we introduce structured knowledge representations ($GU_{json}$), which significantly improve summarization performance by 40% and 14% across two public datasets. Most notably, we propose the Chain-of-Key strategy ($CoK_{json}$) that dynamically updates or augments these representations with new information, rather than recreating the structured memory for each new source. This method further enhances performance by 7% and 4% on the datasets.