GraphER: An Efficient Graph-Based Enrichment and Reranking Method for Retrieval-Augmented Generation

Ruizhong Miao; Yuying Wang; Rongguang Wang; Chenyang Li; Tao Sheng; Sujith Ravi; Dan Roth

cs.LGMar 26, 2026

GraphER: An Efficient Graph-Based Enrichment and Reranking Method for Retrieval-Augmented Generation

Authors: Ruizhong Miao, Yuying Wang, Rongguang Wang, Chenyang Li, Tao Sheng, Sujith Ravi, Dan Roth

Organizations: Oracle AI

Abstract

Semantic search in retrieval-augmented generation (RAG) systems is often insufficient for complex information needs, particularly when relevant evidence is scattered across multiple sources, because it may fail to retrieve the complete set of evidence. Existing approaches to addressing this problem either rely on iterative agentic retrieval, which can be computationally inefficient, or maintain additional structures such as knowledge graphs, which introduce storage and maintenance overhead. In this paper, we propose GraphER, a graph-based enrichment and reranking framework that (1) leverages the organizational structure of data to capture proximity relationships beyond semantic similarity, (2) constructs a graph at query time based on these proximities, and (3) applies graph-based ranking to surface the top candidate documents. Experiments across table retrieval, multi-hop retrieval, and long-document retrieval benchmarks demonstrate consistent improvements in terms of retrieval completeness. Additionally, GraphER requires no additional graph infrastructure and integrates seamlessly with standard vector stores. The framework is retriever-agnostic, supports multiple forms of proximity, and introduces minimal query-time latency.

Figures & tables

Figure 1: Example of RAG for SQL generation: Semantic search misses the Customers table in the top-5 results, whereas GraphER reranks the candidates to retrieve all relevant tables.

Figure 2: During offline indexing, each data object is enriched with metadata that informs edge relationships and then indexed using standard retrieval methods. During online retrieval, a base retriever retrieves candidate objects and assigns initial scores, after which a graph is constructed from the candidates based on the enrichments. A graph-based ranking algorithm is then applied to produce the final relevance scores.

\bm{W}_{ij}=\begin{cases}\frac{\bm{A}_{ij}}{\sum_{k=1}^{n}\bm{A}_{ik}}&\text{if }\bm{A}_{ij}>0\\ 0&\text{otherwise}.\end{cases}

Algorithm 1 Graph Cohesive Smoothing

	Spider1_train	Spider1_dev	Spider1_test	Bird_train	Bird_dev	Beaver_dev	HotpotQA	2WikiMultihopQA	MuSiQue	BEIR-NQ
# queries	8650	1032	2147	9398	1534	209	2000	2000	2417	3452
# queries with $>$ 1 rel. object	3658	378	787	7679	1218	206	2000	2000	2417	666
# data objects	876	876	1056	522	75	463	19323	12601	21100	117683

Table 1: Dataset statistics.

	Dataset
Retriever	Spider1_train	Spider1_dev	Spider1_test	Bird_dev	Beaver_dev
PR@5 (%)
Llama-Embed-Nemotron-8B+BM25	23.7 (42.5)	37.3 (57.3)	22.2 (41.8)	21.8 (30.8)	14.1 (14.8)
Llama-Embed-Nemotron-8B+BM25 $+$ GraphER-PPR	29.7 ( 44.4 )	44.7 (58.8)	26.9 (41.6)	33.0 ( 40.3 )	11.7 (12.4)
Llama-Embed-Nemotron-8B+BM25 $+$ GraphER-GCS	29.4 ( 44.9 )	47.6 ( 61.0 )	29.2 ( 44.6 )	29.4 (36.6)	17.0 ( 17.7 )
Llama-Embed-Nemotron-8B+BM25 $+$ GraphER-GAT	N/A	49.7 ( 62.5 )	40.9 ( 54.4 )	36.9 ( 43.2 )	15.0 ( 15.8 )

Table 2: Retrieval performance on table retrieval benchmarks. We highlight the best and second-best results within each dataset-base retriever setting. Numbers in parentheses denote PR@K over all queries; numbers outside parentheses denote PR@K for queries with more than one relevant object.

PR@5 (%)
	Dataset
Retriever	HotpotQA	2WikiMultihopQA	MuSiQue
Llama-Embed-Nemotron-8B+BM25	78.5	42.6	24.2
Llama-Embed-Nemotron-8B+BM25 $+$ GraphER-PPR	72.9	42.1	17.3
Llama-Embed-Nemotron-8B+BM25 $+$ GraphER-GCS	78.8	43.8	25.4
Llama-Embed-Nemotron-8B+BM25 $+$ GraphER-GAT	78.0	44.1	25.6

Table 3: Retrieval performance on multi-hop QA benchmarks. Numbers in parentheses denote the percentages of queries answered correctly, as evaluated by GPT-5.

PR@5 (%)
	Dataset
Retriever	BEIR-NQ
Llama-Embed-Nemotron-8B+BM25	50.9 (77.4)
Llama-Embed-Nemotron-8B+BM25 $+$ GraphER-PPR	53.3 (76.8)
Llama-Embed-Nemotron-8B+BM25 $+$ GraphER-GCS	52.3 ( 78.1 )
Llama-Embed-Nemotron-8B+BM25 $+$ GraphER-GAT	54.4 ( 78.2 )

Table 4: Retrieval performance on BEIR-NQ. Numbers in parentheses denote PR@K over all queries; numbers outside parentheses denote PR@K for queries with more than one relevant object.

	Dataset
Reranker	Spider1_dev	Spider1_test	Bird_dev	Beaver_dev	HotpotQA	2WikiMultihopQA	MuSiQue	BEIR-NQ
PR@5 (%)
GraphER-GCS	47.6 (61.0)	29.2 (44.6)	29.4 (36.6)	17.0 ( 17.7 )	78.8	43.8	25.4	52.3 (78.1)
GraphER-GAT	49.7 ( 62.5 )	40.9 ( 54.4 )	36.9 ( 43.2 )	15.0 (15.8)	78.0	44.1	25.6	54.4 ( 78.2 )
GraphER-MLP	45.2 (60.5)	36.7 (52.7)	35.2 (42.5)	15.0 (15.8)	78.9	42.5	21.6	50.3 (76.2)
PR@10 (%)

Table 5: Comparison of GraphER-GCS, GraphER-GAT, and GraphER-MLP on evaluation benchmarks. The base retriever is Llama-Embed-Nemotron-8B+BM25.

Multi-Hop Retrieval

See all 16 papers

Corpus-Guided Dual-Path Propagation for Graph Retrieval-Augmented Generation
Sep 29, 2026Baoxian Liu, Tong WeiMulti-Hop QA Graph Retrieval-Augmented Generation
College of Software Engineering, Southeast University, Nanjing 210096, China · Key Laboratory of Computer Network and Information Integration (Southeast University), Ministry of Education, China · School of Computer Science and Engineering, Southeast University, Nanjing 210096, China
Graphs RAG at Scale: Beyond Retrieval-Augmented Generation With Labeled Property Graphs and Resource Description Framework for Complex and Unknown Search Spaces
Mar 21, 2026Manie Tadayon, Mayank GuptaGraph Retrieval-Augmented Generation Graph-Based Retrieval
AI Scientist, Capital Group
EHRAG: Bridging Semantic Gaps in Lightweight GraphRAG via Hybrid Hypergraph Construction and Retrieval
Apr 19, 2026Yifan Song, Xingjian Tao, Zhicheng Yang +2Graph Retrieval-Augmented Generation Graph-Based Retrieval
1The Hong Kong University of Science and Technology (Guangzhou) · 2The Hong Kong University of Science and Technology

Sep 29, 2026J/K move · Enter open · S save

Corpus-Guided Dual-Path Propagation for Graph Retrieval-Augmented Generation

Baoxian Liu, Tong Wei

College of Software Engineering, Southeast University, Nanjing 210096, China · Key Laboratory of Computer Network and Information Integration (Southeast University), Ministry of Education, China · School of Computer Science and Engineering, Southeast University, Nanjing 210096, China

Multi-Hop QA Graph Retrieval-Augmented Generation Graph-Based Retrieval

Open paper AskPDF arXiv

Graph-based retrieval-augmented generation supports multi-hop retrieval by organizing corpus information into graphs. However, existing relation-free graph retrieval methods rely primarily on query-sentence similarity to search for evidence. This can exclude useful bridging evidence with low query similarity and activate incidental entities unrelated to the reasoning chain. In this paper, we propose a simple and effective approach called NexusRAG, which augments the relation-free Tri-Graph with a corpus-level entity neighborhood structure derived from joint entity co-occurrence and semantic similarity. NexusRAG employs this structure to guide two complementary propagation paths: neighborhood-constrained semantic propagation through sentences identifies the query-relevant entity frontier, while direct structural propagation between neighboring entities expands that frontier to structurally related entities. The propagated entity weights also inform neighborhood-aware passage initialization for Personalized PageRank. Experiments on three multi-hop QA benchmarks and a domain-specific subset of GraphRAG-Bench show that NexusRAG consistently outperforms existing approaches. On the GraphRAG-Bench subset, NexusRAG achieves the highest evidence recall in all question categories, exceeding baselines by 4.2-8.1 points. The implementation code is available at https://github.com/Jacob-biu/NexusRAG.

Abstract

Multi-Hop Retrieval

Explore similar work

Corpus-Guided Dual-Path Propagation for Graph Retrieval-Augmented Generation

Graphs RAG at Scale: Beyond Retrieval-Augmented Generation With Labeled Property Graphs and Resource Description Framework for Complex and Unknown Search Spaces

EHRAG: Bridging Semantic Gaps in Lightweight GraphRAG via Hybrid Hypergraph Construction and Retrieval