Rethinking Retrieval-Augmentation as Synthesis: A Query-Aware Context Merging Approach

TL;DR

This paper introduces MergeRAG, a novel retrieval-augmentation framework that dynamically synthesizes retrieved contexts to improve information density and model performance within the limited context window of LLMs.

Contribution

MergeRAG shifts retrieval-augmentation from static filtering to query-aware synthesis using a dual-pathway merging strategy and hierarchical parallel merging, significantly enhancing retrieval effectiveness.

Findings

Achieves up to 13.7 points improvement in F1 score.

Achieves up to 11.5 points improvement in Exact Match.

Outperforms state-of-the-art RAG baselines on standard benchmarks.

Abstract

Retrieval-Augmented Generation (RAG) enables Large Language Models (LLMs) to extend their existing knowledge by dynamically incorporating external information. However, practical deployment is fundamentally constrained by the LLM's finite context window, forcing a trade-off between information sufficiency and token consumption. Standard pipelines address this via a retrieve-then-select strategy, typically retaining only the top-k chunks based on relevance. Nevertheless, this approach is suboptimal: it inherently truncates critical bridging evidence located in the long tail of the relevance distribution, while simultaneously wasting the token budget on semantically redundant high-ranking chunks. In this paper, we rethink retrieval-augmentation as a dynamic optimization problem aimed at maximizing information density. We propose MergeRAG, a novel framework that shifts the paradigm from static filtering to query-aware synthesis. MergeRAG employs a scoring agent to restructure retrieved contexts through a dual-pathway mechanism: 1) Symmetric Merging, which consolidates weak signals to recover lost bridging evidence; 2) Asymmetric Merging, which utilizes entropy-guided anchoring to eliminate redundancy without sacrificing semantic integrity. We further introduce a Hierarchical Parallel Merging strategy that mitigates information loss while maximizing computational parallelism. Extensive experiments on standard benchmarks demonstrate that MergeRAG significantly outperforms state-of-the-art RAG baselines, achieving up to 13.7 points improvement in F1 score and 11.5 points in Exact Match (EM), respectively.