LLM-Specific Utility: A New Perspective for Retrieval-Augmented Generation

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Zhang, Hengran, Bi, Keping, Guo, Jiafeng, Zhang, Jiaming, Wang, Shuaiqiang, Yin, Dawei, Cheng, Xueqi
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866918307113730048
author Zhang, Hengran
Bi, Keping
Guo, Jiafeng
Zhang, Jiaming
Wang, Shuaiqiang
Yin, Dawei
Cheng, Xueqi
author_facet Zhang, Hengran
Bi, Keping
Guo, Jiafeng
Zhang, Jiaming
Wang, Shuaiqiang
Yin, Dawei
Cheng, Xueqi
contents Retrieval-augmented generation (RAG) is typically optimized for topical relevance, yet its success ultimately depends on whether retrieved passages are useful for a large language model (LLM) to generate correct and complete answers. We argue that such utility is often LLM-specific rather than universal, due to differences in models' knowledge, reasoning, and ability to leverage evidence. We formalize LLM-specific utility as the performance improvement of a target LLM when a passage is provided, compared to answering without evidence. To systematically study LLM-specific utility, we construct a benchmark of LLM-specific gold utilitarian passages for four LLMs (Qwen3-8B/14B/32B and Llama3.1-8B) on three QA datasets (Natural Questions, TriviaQA, and MS MARCO-FQA). Our analysis shows that utilitarian passages are model-dependent and non-transferable: each LLM performs best with its own utilitarian evidence, while evidence optimized for other LLMs is consistently suboptimal. Human-annotated evidence remains a strong general baseline but does not fully match individual LLM utility needs. We further introduce the LLM-specific utility judgment task and find that existing utility-aware selection and scoring methods largely capture model-agnostic usefulness and struggle to reliably estimate LLM-specific utility. Overall, our findings highlight the limitations of current utility-aware retrieval and motivate generator-tailored evidence selection for improving RAG.
format Preprint
id arxiv_https___arxiv_org_abs_2510_11358
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LLM-Specific Utility: A New Perspective for Retrieval-Augmented Generation
Zhang, Hengran
Bi, Keping
Guo, Jiafeng
Zhang, Jiaming
Wang, Shuaiqiang
Yin, Dawei
Cheng, Xueqi
Computation and Language
Artificial Intelligence
Information Retrieval
Retrieval-augmented generation (RAG) is typically optimized for topical relevance, yet its success ultimately depends on whether retrieved passages are useful for a large language model (LLM) to generate correct and complete answers. We argue that such utility is often LLM-specific rather than universal, due to differences in models' knowledge, reasoning, and ability to leverage evidence. We formalize LLM-specific utility as the performance improvement of a target LLM when a passage is provided, compared to answering without evidence. To systematically study LLM-specific utility, we construct a benchmark of LLM-specific gold utilitarian passages for four LLMs (Qwen3-8B/14B/32B and Llama3.1-8B) on three QA datasets (Natural Questions, TriviaQA, and MS MARCO-FQA). Our analysis shows that utilitarian passages are model-dependent and non-transferable: each LLM performs best with its own utilitarian evidence, while evidence optimized for other LLMs is consistently suboptimal. Human-annotated evidence remains a strong general baseline but does not fully match individual LLM utility needs. We further introduce the LLM-specific utility judgment task and find that existing utility-aware selection and scoring methods largely capture model-agnostic usefulness and struggle to reliably estimate LLM-specific utility. Overall, our findings highlight the limitations of current utility-aware retrieval and motivate generator-tailored evidence selection for improving RAG.
title LLM-Specific Utility: A New Perspective for Retrieval-Augmented Generation
topic Computation and Language
Artificial Intelligence
Information Retrieval
url https://arxiv.org/abs/2510.11358