From Sparse to Dense: GPT-4 Summarization with Chain of Density Prompting

TL;DR

This paper introduces a 'Chain of Density' prompting method for GPT-4 that iteratively creates summaries with optimal density, balancing informativeness and readability, and demonstrates its effectiveness through human preference studies.

Contribution

It proposes a novel iterative prompting technique, Chain of Density, to generate more informative and abstractive summaries without increasing length, improving over vanilla GPT-4 prompts.

Findings

GPT-4 with CoD produces more abstractive summaries.

Humans prefer denser GPT-4 summaries over vanilla prompts.

CoD summaries are nearly as dense as human-written summaries.

Abstract

Selecting the ``right'' amount of information to include in a summary is a difficult task. A good summary should be detailed and entity-centric without being overly dense and hard to follow. To better understand this tradeoff, we solicit increasingly dense GPT-4 summaries with what we refer to as a ``Chain of Density'' (CoD) prompt. Specifically, GPT-4 generates an initial entity-sparse summary before iteratively incorporating missing salient entities without increasing the length. Summaries generated by CoD are more abstractive, exhibit more fusion, and have less of a lead bias than GPT-4 summaries generated by a vanilla prompt. We conduct a human preference study on 100 CNN DailyMail articles and find that that humans prefer GPT-4 summaries that are more dense than those generated by a vanilla prompt and almost as dense as human written summaries. Qualitative analysis supports the notion that there exists a tradeoff between informativeness and readability. 500 annotated CoD summaries, as well as an extra 5,000 unannotated summaries, are freely available on HuggingFace (https://huggingface.co/datasets/griffin/chain_of_density).