SARA: Selective and Adaptive Retrieval-augmented Generation with Context Compression

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jin, Yiqiao, Sharma, Kartik, Rakesh, Vineeth, Dou, Yingtong, Pan, Menghai, Das, Mahashweta, Kumar, Srijan
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916832097599488
author Jin, Yiqiao
Sharma, Kartik
Rakesh, Vineeth
Dou, Yingtong
Pan, Menghai
Das, Mahashweta
Kumar, Srijan
author_facet Jin, Yiqiao
Sharma, Kartik
Rakesh, Vineeth
Dou, Yingtong
Pan, Menghai
Das, Mahashweta
Kumar, Srijan
contents Retrieval-augmented Generation (RAG) extends large language models (LLMs) with external knowledge but faces key challenges: restricted effective context length and redundancy in retrieved documents. Pure compression-based approaches reduce input size but often discard fine-grained details essential for factual accuracy. We propose SARA, a unified RAG framework that balances local precision and global knowledge coverage under tight context budgets. SARA combines natural-language text snippets with semantic compression vectors to jointly enhance context efficiency and answer correctness. It represents contexts at two complementary levels: 1) fine-grained natural-language spans that preserve critical entities and numerical values, and 2) compact, interpretable vectors that summarize high-level semantics. An iterative evidence-selection module employs the compression vectors for dynamic reranking of contexts. Across 9 datasets and 5 open-source LLMs spanning 3 model families (Mistral, Llama, and Gemma), SARA consistently improves answer relevance (+17.71), answer correctness (+13.72), and semantic similarity (+15.53), demonstrating the importance of integrating textual and compressed representations for robust, context-efficient RAG.
format Preprint
id arxiv_https___arxiv_org_abs_2507_05633
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SARA: Selective and Adaptive Retrieval-augmented Generation with Context Compression
Jin, Yiqiao
Sharma, Kartik
Rakesh, Vineeth
Dou, Yingtong
Pan, Menghai
Das, Mahashweta
Kumar, Srijan
Computation and Language
Artificial Intelligence
Information Retrieval
Retrieval-augmented Generation (RAG) extends large language models (LLMs) with external knowledge but faces key challenges: restricted effective context length and redundancy in retrieved documents. Pure compression-based approaches reduce input size but often discard fine-grained details essential for factual accuracy. We propose SARA, a unified RAG framework that balances local precision and global knowledge coverage under tight context budgets. SARA combines natural-language text snippets with semantic compression vectors to jointly enhance context efficiency and answer correctness. It represents contexts at two complementary levels: 1) fine-grained natural-language spans that preserve critical entities and numerical values, and 2) compact, interpretable vectors that summarize high-level semantics. An iterative evidence-selection module employs the compression vectors for dynamic reranking of contexts. Across 9 datasets and 5 open-source LLMs spanning 3 model families (Mistral, Llama, and Gemma), SARA consistently improves answer relevance (+17.71), answer correctness (+13.72), and semantic similarity (+15.53), demonstrating the importance of integrating textual and compressed representations for robust, context-efficient RAG.
title SARA: Selective and Adaptive Retrieval-augmented Generation with Context Compression
topic Computation and Language
Artificial Intelligence
Information Retrieval
url https://arxiv.org/abs/2507.05633