Saved in:
Bibliographic Details
Main Authors: Koshorek, Omri, Granot, Niv, Alloni, Aviv, Admati, Shahar, Hendel, Roee, Weiss, Ido, Arazi, Alan, Cohen, Shay-Nitzan, Belinkov, Yonatan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2511.08505
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908644831920128
author Koshorek, Omri
Granot, Niv
Alloni, Aviv
Admati, Shahar
Hendel, Roee
Weiss, Ido
Arazi, Alan
Cohen, Shay-Nitzan
Belinkov, Yonatan
author_facet Koshorek, Omri
Granot, Niv
Alloni, Aviv
Admati, Shahar
Hendel, Roee
Weiss, Ido
Arazi, Alan
Cohen, Shay-Nitzan
Belinkov, Yonatan
contents Retrieval-Augmented Generation (RAG) has become the dominant approach for answering questions over large corpora. However, current datasets and methods are highly focused on cases where only a small part of the corpus (usually a few paragraphs) is relevant per query, and fail to capture the rich world of aggregative queries. These require gathering information from a large set of documents and reasoning over them. To address this gap, we propose S-RAG, an approach specifically designed for such queries. At ingestion time, S-RAG constructs a structured representation of the corpus; at inference time, it translates natural-language queries into formal queries over said representation. To validate our approach and promote further research in this area, we introduce two new datasets of aggregative queries: HOTELS and WORLD CUP. Experiments with S-RAG on the newly introduced datasets, as well as on a public benchmark, demonstrate that it substantially outperforms both common RAG systems and long-context LLMs.
format Preprint
id arxiv_https___arxiv_org_abs_2511_08505
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Structured RAG for Answering Aggregative Questions
Koshorek, Omri
Granot, Niv
Alloni, Aviv
Admati, Shahar
Hendel, Roee
Weiss, Ido
Arazi, Alan
Cohen, Shay-Nitzan
Belinkov, Yonatan
Computation and Language
Machine Learning
Retrieval-Augmented Generation (RAG) has become the dominant approach for answering questions over large corpora. However, current datasets and methods are highly focused on cases where only a small part of the corpus (usually a few paragraphs) is relevant per query, and fail to capture the rich world of aggregative queries. These require gathering information from a large set of documents and reasoning over them. To address this gap, we propose S-RAG, an approach specifically designed for such queries. At ingestion time, S-RAG constructs a structured representation of the corpus; at inference time, it translates natural-language queries into formal queries over said representation. To validate our approach and promote further research in this area, we introduce two new datasets of aggregative queries: HOTELS and WORLD CUP. Experiments with S-RAG on the newly introduced datasets, as well as on a public benchmark, demonstrate that it substantially outperforms both common RAG systems and long-context LLMs.
title Structured RAG for Answering Aggregative Questions
topic Computation and Language
Machine Learning
url https://arxiv.org/abs/2511.08505