跳到正文
arXiv cs.CL· Taegyeong Lee, Jiwon Park, Seunghyun Hwang, JooYoung Jang·· 10 小时前AI 评分26

DEO:面向否定感知检索的免训练直接嵌入优化方法

DEO: Training-Free Direct Embedding Optimization for Negation-Aware Retrieval

AI 导读

研究者提出免训练方法 DEO,将查询拆分为正向与负向成分并用对比目标优化查询嵌入,无需额外训练数据或模型更新即可处理否定与排除类检索。在 NegConstraint 上,DEO 较基线提升 +0.0738 nDCG@10 和 +0.1028 MAP@100;多模态检索中 Recall@5 较 OpenAI CLIP 提升 +6%。

正文 · 原文

View PDF HTML (experimental)

Abstract:Recent advances in Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) have enabled diverse retrieval methods. However, existing retrieval methods often fail to accurately retrieve results for negation and exclusion queries. To address this limitation, prior approaches rely on embedding adaptation or fine-tuning, which introduce additional computational cost and deployment complexity. We propose Direct Embedding Optimization (DEO), a training-free method for negation-aware text and multimodal retrieval. DEO decomposes queries into positive and negative components and optimizes the query embedding with a contrastive objective. Without additional training data or model updates, DEO outperforms baselines on NegConstraint, with gains of +0.0738 nDCG@10 and +0.1028 MAP@100, while improving Recall@5 by +6\% over OpenAI CLIP in multimodal retrieval. These results demonstrate the practicality of DEO for negation- and exclusion-aware retrieval in real-world settings.
Comments: Accepted to the RAG4Report Workshop at ACL 2026. Fixed typos and added acknowledgments
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2603.09185 [cs.CL]
  (or arXiv:2603.09185v2 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2603.09185

arXiv-issued DOI via DataCite

Submission history

From: Taegyeong Lee [view email]
[v1] Tue, 10 Mar 2026 04:47:15 UTC (1,315 KB)
[v2] Thu, 19 Nov 2026 07:58:50 UTC (1,315 KB)

来源:arXiv cs.CL · arxiv.org