Enhancing Incremental Summarization with Structured Representations
This addresses the issue of information overload in incremental summarization for users of LLMs, though it appears incremental as it builds on existing methods with structured representations.
The paper tackles the problem of LLMs struggling with extensive input contexts in incremental summarization by introducing structured knowledge representations, which improve performance by 40% and 14% on two datasets, and a Chain-of-Key strategy that further enhances it by 7% and 4%.
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.