A Unified Retrieval Framework with Document Ranking and EDU Filtering for Multi-document Summarization

TL;DR

The paper addresses the context-length constraint in multi-document summarization by proposing ReREF, a retrieval framework that unifies latent EDU query selection, document ranking, and EDU filtering via an EM optimization process. EDUs are extracted and embedded (via token, EDU, and document representations), and latent queries are dynamically inferred to guide ranking and filtering, with the objective $${L}_{\text{total}} = {L}_{\text{rank}} + \lambda {L}_{\text{filter}}$$. Across four diverse MDS datasets and multiple base models, ReREF yields consistent ROUGE improvements and demonstrates robustness in few-shot settings, with additional human evaluations indicating gains in informativeness and fluency. The approach is model-agnostic and scalable, offering a practical solution to preserving critical information under strict context-length limits and reducing irrelevant content in multi-document inputs.

Abstract

In the field of multi-document summarization (MDS), transformer-based models have demonstrated remarkable success, yet they suffer an input length limitation. Current methods apply truncation after the retrieval process to fit the context length; however, they heavily depend on manually well-crafted queries, which are impractical to create for each document set for MDS. Additionally, these methods retrieve information at a coarse granularity, leading to the inclusion of irrelevant content. To address these issues, we propose a novel retrieval-based framework that integrates query selection and document ranking and shortening into a unified process. Our approach identifies the most salient elementary discourse units (EDUs) from input documents and utilizes them as latent queries. These queries guide the document ranking by calculating relevance scores. Instead of traditional truncation, our approach filters out irrelevant EDUs to fit the context length, ensuring that only critical information is preserved for summarization. We evaluate our framework on multiple MDS datasets, demonstrating consistent improvements in ROUGE metrics while confirming its scalability and flexibility across diverse model architectures. Additionally, we validate its effectiveness through an in-depth analysis, emphasizing its ability to dynamically select appropriate queries and accurately rank documents based on their relevance scores. These results demonstrate that our framework effectively addresses context-length constraints, establishing it as a robust and reliable solution for MDS.

Figures (7)

• Figure 1: (A) The overall procedure of our retrieval model, which is model agnostic. (B) Previous retrieval needs queries as inputs and ranking over documents. (C) Our retrieval automatically selects queries for document ranking and applies filtering over EDUs.
• Figure 2: Example EDUs, segmented by DMRST liu-etal-2021-dmrst, demonstrate their finer granularity than sentences while being more semantically meaningful than words or tokens.
• Figure 3: The overall process of ReREF, which extracts latent queries and performs ranking and filtering simultaneously.
• Figure 4: Precision@K (K = 10, 20, 50, 100) for query selection and filtering. (A) (B): the proportion of the top-K selected EDUs for query selection that align with the most salient EDUs. (C) (D): the proportion of the bottom-K selected EDUs for filtering that align with the least salient EDUs.
• Figure 5: Document ranking accuracy. NDCG@K evaluates how effectively the most relevant documents are ranked within the top K positions. MRR_1st and MRR_2rd measure the average rank accuracy of the most and second most relevant documents, respectively. (A): Multi-News. (B): WCEP-10.