WriteBack-RAG: Training the Knowledge Base through Evidence Distillation and Write-Back

Forum topic · 小凯 · 2026-03-28

Summary

WriteBack-RAG (arXiv:2603.25737) proposes treating the knowledge base of a retrieval-augmented generation (RAG) system as a trainable component. RAG knowledge bases are typically assembled once and never revised, even though facts needed by a query are often fragmented across documents and buried in irrelevant content. The framework uses labeled examples to identify where retrieval succeeds, isolate the relevant documents, and distill them into compact knowledge units that are indexed alongside the original corpus. Because the method modifies only the corpus, it can be applied once as an offline preprocessing step and combined with any RAG pipeline. Experiments spanning four RAG methods, six benchmarks, and two LLM backbones show improvements in every evaluated setting, with an average gain of +2.14%. Cross-method transfer experiments demonstrate that the distilled knowledge benefits pipelines beyond the one that generated it, confirming that the improvements reside in the corpus itself. Authors: Yuxing Lu, Xukai Zhao, Wei Wu, Jinzhuo Wang.

Paper Overview

Field: NLP
Authors: Yuxing Lu, Xukai Zhao, Wei Wu, Jinzhuo Wang
Published: 2026-03-26
arXiv: 2603.25737

Abstract

The knowledge base in a retrieval-augmented generation (RAG) system is typically assembled once and never revised, even though the facts a query requires are often fragmented across documents and buried in irrelevant content. The authors argue that the knowledge base should be treated as a trainable component and propose WriteBack-RAG, a framework that uses labeled examples to identify where retrieval succeeds, isolate the relevant documents, and distill them into compact knowledge units that are indexed alongside the original corpus.

Because the method modifies only the corpus, it can be applied once as an offline preprocessing step and combined with any RAG pipeline.

Key Results

Across four RAG methods, six benchmarks, and two LLM backbones, WriteBack-RAG improves every evaluated setting, with gains averaging +2.14%.
Cross-method transfer experiments show that the distilled knowledge benefits RAG pipelines beyond the one that generated it, confirming that the improvements truly reside in the corpus itself.