SPD-RAG: Sub-Agent Per Document Retrieval-Augmented Generation
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866910046363844608 |
|---|---|
| author | Akay, Yagiz Can Kartal, Muhammed Yusuf Alparslan, Esra Ortakoyluoglu, Faruk Akpinar, Arda |
| author_facet | Akay, Yagiz Can Kartal, Muhammed Yusuf Alparslan, Esra Ortakoyluoglu, Faruk Akpinar, Arda |
| contents | Answering complex, real-world queries often requires synthesizing facts scattered across vast document corpora. In these settings, standard retrieval-augmented generation (RAG) pipelines suffer from incomplete evidence coverage, while long-context large language models (LLMs) struggle to reason reliably over massive inputs. We introduce SPD-RAG, a hierarchical multi-agent framework for exhaustive cross-document question answering that decomposes the problem along the document axis. Each document is processed by a dedicated document-level agent operating only on its own content, enabling focused retrieval, while a coordinator dispatches tasks to relevant agents and aggregates their partial answers. Agent outputs are synthesized by merging partial answers through a token-bounded synthesis layer (which supports recursive map-reduce for massive corpora). This document-level specialization with centralized fusion improves scalability and answer quality in heterogeneous multidocument settings while yielding a modular, extensible retrieval pipeline. On the LOONG benchmark (EMNLP 2024) for long-context multi-document QA, SPD-RAG achieves an Avg Score of 58.1 (GPT-5 evaluation), outperforming Normal RAG (33.0) and Agentic RAG (32.8) while using only 38% of the API cost of a full-context baseline (68.0). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2603_08329 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | SPD-RAG: Sub-Agent Per Document Retrieval-Augmented Generation Akay, Yagiz Can Kartal, Muhammed Yusuf Alparslan, Esra Ortakoyluoglu, Faruk Akpinar, Arda Computation and Language Artificial Intelligence Information Retrieval 68T50, 68T42, 68P20 (Primary) H.3.3; H.3.4; I.2.11; I.2.7 Answering complex, real-world queries often requires synthesizing facts scattered across vast document corpora. In these settings, standard retrieval-augmented generation (RAG) pipelines suffer from incomplete evidence coverage, while long-context large language models (LLMs) struggle to reason reliably over massive inputs. We introduce SPD-RAG, a hierarchical multi-agent framework for exhaustive cross-document question answering that decomposes the problem along the document axis. Each document is processed by a dedicated document-level agent operating only on its own content, enabling focused retrieval, while a coordinator dispatches tasks to relevant agents and aggregates their partial answers. Agent outputs are synthesized by merging partial answers through a token-bounded synthesis layer (which supports recursive map-reduce for massive corpora). This document-level specialization with centralized fusion improves scalability and answer quality in heterogeneous multidocument settings while yielding a modular, extensible retrieval pipeline. On the LOONG benchmark (EMNLP 2024) for long-context multi-document QA, SPD-RAG achieves an Avg Score of 58.1 (GPT-5 evaluation), outperforming Normal RAG (33.0) and Agentic RAG (32.8) while using only 38% of the API cost of a full-context baseline (68.0). |
| title | SPD-RAG: Sub-Agent Per Document Retrieval-Augmented Generation |
| topic | Computation and Language Artificial Intelligence Information Retrieval 68T50, 68T42, 68P20 (Primary) H.3.3; H.3.4; I.2.11; I.2.7 |
| url | https://arxiv.org/abs/2603.08329 |