Coverage-based Fairness in Multi-document Summarization

TL;DR

This paper introduces new fairness measures for multi-document summarization that account for redundancy and corpus-level disparities, and evaluates various language models using these measures.

Contribution

It proposes the Equal Coverage and Coverage Parity measures, addressing limitations of previous fairness metrics in MDS, and applies them to evaluate multiple LLMs.

Findings

Claude3-sonnet is the fairest LLM evaluated.

Most LLMs tend to overrepresent certain social attributes.

The new measures align better with fairness definitions.

Abstract

Fairness in multi-document summarization (MDS) measures whether a system can generate a summary fairly representing information from documents with different social attribute values. Fairness in MDS is crucial since a fair summary can offer readers a comprehensive view. Previous works focus on quantifying summary-level fairness using Proportional Representation, a fairness measure based on Statistical Parity. However, Proportional Representation does not consider redundancy in input documents and overlooks corpus-level unfairness. In this work, we propose a new summary-level fairness measure, Equal Coverage, which is based on coverage of documents with different social attribute values and considers the redundancy within documents. To detect the corpus-level unfairness, we propose a new corpus-level measure, Coverage Parity. Our human evaluations show that our measures align more with our definition of fairness. Using our measures, we evaluate the fairness of thirteen different LLMs. We find that Claude3-sonnet is the fairest among all evaluated LLMs. We also find that almost all LLMs overrepresent different social attribute values. The code is available at https://github.com/leehaoyuan/coverage_fairness.