arXiv：cs.CL· Aditya Agrawal, Alwarappan Nakkiran, Aman Singh Thakur, Alex Karlsson, Harsha Aduri·· 3 小时前AI 评分43
RAG 必须超越事实性 grounding，以表征多元意见
Retrieval-Augmented Generation Must Move Beyond Factual Grounding to Represent Diverse Opinions
AI 导读
一篇立场论文指出，RAG 系统默认查询存在正确答案、检索应收敛于此，由此产生事实性偏差，忽视意见型内容中的随机不确定性，带来少数声音被抹除和意见操纵的风险。作者用不确定性量化形式化意见感知检索，并提出基于 Wasserstein 距离的统一目标与 Opinion-Aware RAG（O-RAG），在文档索引前用 LLM 抽取实体关联的意见元数据。
正文
Abstract:Retrieval-Augmented Generation (RAG) systems are built on an unexamined assumption - that queries have correct answers and retrieval should converge toward them. This position paper argues that this creates a factual bias where RAG systems optimize for reducing epistemic uncertainty while ignoring the aleatoric uncertainty, inherent in opinion-rich content. The consequences go beyond technical limitations- due to risk of minority voice erasure and risk of opinion manipulation. To address this, we formalize opinion-aware retrieval through uncertainty quantification and derive a unified objective using the Wasserstein distance. As an existence proof, we present Opinion-Aware RAG (O-RAG), which enriches documents with LLM-extracted, entity-linked opinion metadata before indexing. Across e-commerce seller forums and public hotel reviews, O-RAG reduces Wasserstein distance to corpus-level sentiment distributions by 18-48%, and human evaluators preferred its responses 79.2% of the time. We close with a research agenda for opinion-aware RAG.
| Comments: | 17 pages, Accepted at 19th International Conference on Natural Language Generation 2026 |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Information Retrieval (cs.IR) |
| Cite as: | arXiv:2604.12138 [cs.AI] |
| (or arXiv:2604.12138v5 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2604.12138 arXiv-issued DOI via DataCite |
Submission history
From: Aditya Agrawal [view email]
[v1]
Mon, 13 Apr 2026 23:39:39 UTC (59 KB)
[v2]
Wed, 3 Jun 2026 18:44:26 UTC (33 KB)
[v3]
Wed, 8 Jul 2026 18:30:50 UTC (48 KB)
[v4]
Fri, 2 Oct 2026 19:37:05 UTC (47 KB)
[v5]
Tue, 6 Oct 2026 17:18:54 UTC (49 KB)
来源：arXiv：cs.CL · arxiv.org