| Challenge: | Existing studies quantify summary-level fairness using Proportional Representation, but they ignore corpus-level unfairness. |
| Approach: | They propose a new summary-level fairness measure that considers redundancy in documents . they evaluate the fairness of thirteen different multi-document summarization systems . |
| Outcome: | The proposed measure is based on coverage of documents with different social attribute values and considers redundancy within documents. |
Similar Papers
Improving Fairness of Large Language Models in Multi-document Summarization (2025.acl-short)
Copied to clipboard
| Challenge: | Recent studies focus on summary-level fairness, while corpus-level focuses on corpus of summaries. |
| Approach: | They propose a preference tuning method that focuses on both summary-level and corpus-level fairness in MDS. |
| Outcome: | The proposed method outperforms baselines while maintaining critical qualities of summaries. |
Corpora Evaluation and System Bias Detection in Multi-document Summarization (2020.findings-emnlp)
Copied to clipboard
| Challenge: | Multi-document summarization (MDS) is a task of combining multiple documents into a concise text paragraph. |
| Approach: | They propose to use a multi-document summarization task to reflect key points from any set of documents into a concise text paragraph. |
| Outcome: | The proposed system performs better on a set of selected datasets than on the other ones. |
Multi Document Summarization Evaluation in the Presence of Damaging Content (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Existing metrics evaluate a summary based on relevance and consistency with the source documents. |
| Approach: | They propose to measure the ability of MDS systems to handle damaging documents in their input set by lexical similarity and language model likelihood. |
| Outcome: | The proposed metrics show that they can summarize a set of documents without damaging content. |
Fair Abstractive Summarization of Diverse Perspectives (2024.naacl-long)
Copied to clipboard
Yusen Zhang, Nan Zhang, Yixin Liu, Alexander Fabbri, Junru Liu, Ryo Kamoi, Xiaoxin Lu, Caiming Xiong, Jieyu Zhao, Dragomir Radev, Kathleen McKeown, Rui Zhang
| Challenge: | Existing work on summarization metrics and large language models has not explored fair abstractive summarizing. |
| Approach: | They propose four reference-free automatic metrics to measure the differences between target and source perspectives. |
| Outcome: | The proposed methods alleviate fair abstractive summarization on user-generated data. |
Open Domain Multi-document Summarization: A Comprehensive Study of Model Brittleness under Retrieval (2023.findings-emnlp)
Copied to clipboard
| Challenge: | Multi-document summarization (MDS) assumes a set of topic-related documents is provided as input. |
| Approach: | They formalize the task and bootstrap it using existing datasets, retrievers and summarizers. |
| Outcome: | The proposed method reduces the sensitivity of summarizers to imperfect retrieval, but is highly sensitive to other errors. |
A Multi-Document Coverage Reward for RELAXed Multi-Document Summarization (2022.acl-long)
Copied to clipboard
| Challenge: | Multi-document summarization models are limited by limited references and with maximum-likelihood objectives. |
| Approach: | They propose to fine-tune an MDS baseline with a reward that balances a reference-based metric such as ROUGE with coverage of the input documents. |
| Outcome: | The proposed model improves on the Multi-News and WCEP datasets with a low-variance estimator . the proposed model also improves the coverage of the input documents . |
How “Multi” is Multi-Document Summarization? (2022.emnlp-main)
Copied to clipboard
| Challenge: | Multi-document summarization (MDS) aims at combining information spread across multiple documents . a single document often covers the full summary content . |
| Approach: | They propose a measure to evaluate the degree to which a summary is "disperse" they propose to combine information from multiple documents into a single document to generate a concise summary . |
| Outcome: | The proposed measure evaluates the degree to which a summary is "disperse" the measure is applied to several popular MDS datasets and state-of-the-art systems. |
Which Information Matters? Dissecting Human-written Multi-document Summaries with Partial Information Decomposition (2024.findings-acl)
Copied to clipboard
| Challenge: | Existing methods to characterize human-written summaries do not account for the nature of high-quality summary. |
| Approach: | They propose to characterize human-written summaries using partial information decomposition . they propose to decompose mutual information provided by all source documents into union, redundancy, synergy, and unique information . |
| Outcome: | The proposed approach decomposes the mutual information provided by all source documents into union, redundancy, synergy, and unique information. |
When Bigger Isn’t Better: A Comprehensive Fairness Evaluation of Political Bias in Multi-News Summarisation (2026.acl-long)
Copied to clipboard
| Challenge: | Existing models that deal with multiple sources can exhibit political biases, causing unequal representation of viewpoints and underrepresentation of minority voices. |
| Approach: | They examine how large language models handle sources with varying political leanings using a dataset with political orientation labels. |
| Outcome: | The proposed model outperforms larger models and offers the best balance of fairness and efficiency. |
The Impossibility of Fair LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Existing frameworks for evaluating large language models do not extend to general-purpose AI contexts or are infeasible in practice. |
| Approach: | They analyze a variety of technical fairness frameworks to find inherent challenges . they find that each framework does not logically extend to the general-purpose AI context . |
| Outcome: | The proposed frameworks do not logically extend to the general-purpose AI context or are infeasible in practice due to large amounts of unstructured training data and potential combinations of human populations, use cases, and sensitive attributes. |