Fair Abstractive Summarization of Diverse Perspectives

TL;DR

This paper addresses the challenge of fair abstractive summarization by defining fairness as balanced representation of diverse perspectives, proposing new metrics, evaluating multiple LLMs, and suggesting methods to improve fairness.

Contribution

It introduces formal fairness metrics for abstractive summarization, evaluates state-of-the-art LLMs on diverse datasets, and proposes techniques to enhance fairness in summaries.

Findings

Both models and human references show low fairness in summaries.

Fairness is influenced by source perspective distribution and model biases.

Proposed methods improve fairness without significantly degrading summary quality.

Abstract

People from different social and demographic groups express diverse perspectives and conflicting opinions on a broad set of topics such as product reviews, healthcare, law, and politics. A fair summary should provide a comprehensive coverage of diverse perspectives without underrepresenting certain groups. However, current work in summarization metrics and Large Language Models (LLMs) evaluation has not explored fair abstractive summarization. In this paper, we systematically investigate fair abstractive summarization for user-generated data. We first formally define fairness in abstractive summarization as not underrepresenting perspectives of any groups of people, and we propose four reference-free automatic metrics by measuring the differences between target and source perspectives. We evaluate nine LLMs, including three GPT models, four LLaMA models, PaLM 2, and Claude, on six datasets collected from social media, online reviews, and recorded transcripts. Experiments show that both the model-generated and the human-written reference summaries suffer from low fairness. We conduct a comprehensive analysis of the common factors influencing fairness and propose three simple but effective methods to alleviate unfair summarization. Our dataset and code are available at https://github.com/psunlpgroup/FairSumm.