A Hybrid Retrieval and Reranking Framework for Evidence-Grounded Retrieval-Augmented Generation

Retrieval-augmented generation (RAG) improves the reliability of large language model outputs by grounding generated responses in external evidence. However, the quality of RAG-generated answers depends strongly on the relevance of retrieved evidence, the effectiveness of evidence ranking, and the ability to verify whether generated claims are supported by source documents. This study presents a hybrid retrieval and reranking framework for citation-aware RAG in biomedical and healthcare-related document question answering. The proposed framework uses Amazon Bedrock Knowledge Bases for document ingestion, parsing, chunking, embedding generation, and evidence retrieval. Source PDF documents are stored in Amazon S3, embedded using Amazon Titan Text Embeddings V2, and indexed using Amazon OpenSearch Serverless. A hybrid retrieval strategy is applied to retrieve candidate evidence chunks, followed by reranking to prioritize the most relevant passages before answer generation. The answer-generation stage uses the highest-ranked evidence chunks to produce controlled, evidence-grounded responses, while a separate judge model evaluates each generated factual claim against the retrieved evidence. The framework was evaluated using 150 biomedical natural language processing and healthcare transformer queries. This evaluation was designed as a pilot-scale proof-of-concept study to examine the feasibility of the proposed hybrid retrieval, reranking, answer-generation, and grounding-evaluation workflow. Across the evaluation set, the system retrieved and reranked 3,000 evidence chunks and generated responses from the top-ranked evidence. Claim-level grounding evaluation identified 1,200 factual claims, of which 1,184 were judged to be supported by the retrieved evidence and 16 were classified as unsupported, resulting in an overall grounding accuracy of 98.7%. The results indicate that hybrid retrieval, reranking, conservative prompting, and claim-level grounding evaluation can support reliable evidence-grounded RAG responses when sufficient source evidence is available. The study also highlights the distinction between internal grounding accuracy and strict citation accuracy, motivating future work on explicit sentence-level citation generation and expanded human expert validation.

Authors

Institutions

Publication Details

Journal
Cureus Journal of Computer Science.
Published
2026-11-18
DOI
https://doi.org/10.7759/s44389-026-00377-4
Citations
2
Primary Topic
Topic Modeling
Type
article
Field-Weighted Citation Impact
1.94
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Hybrid Retrieval and Reranking Framework for Evidence-Grounded Retrieval-Augmented Generation

Sampson Akwafuo, Fariba Afrin Irany
2 citations
Cureus Journal of Computer Science.
Topic Modeling
1.94
article

A Hybrid Retrieval and Reranking Framework for Evidence-Grounded Retrieval-Augmented Generation

Sampson Akwafuo, Fariba Afrin Irany
article en
2 citations

Abstract

Retrieval-augmented generation (RAG) improves the reliability of large language model outputs by grounding generated responses in external evidence. However, the quality of RAG-generated answers depends strongly on the relevance of retrieved evidence, the effectiveness of evidence ranking, and the ability to verify whether generated claims are supported by source documents. This study presents a hybrid retrieval and reranking framework for citation-aware RAG in biomedical and healthcare-related document question answering. The proposed framework uses Amazon Bedrock Knowledge Bases for document ingestion, parsing, chunking, embedding generation, and evidence retrieval. Source PDF documents are stored in Amazon S3, embedded using Amazon Titan Text Embeddings V2, and indexed using Amazon OpenSearch Serverless. A hybrid retrieval strategy is applied to retrieve candidate evidence chunks, followed by reranking to prioritize the most relevant passages before answer generation. The answer-generation stage uses the highest-ranked evidence chunks to produce controlled, evidence-grounded responses, while a separate judge model evaluates each generated factual claim against the retrieved evidence. The framework was evaluated using 150 biomedical natural language processing and healthcare transformer queries. This evaluation was designed as a pilot-scale proof-of-concept study to examine the feasibility of the proposed hybrid retrieval, reranking, answer-generation, and grounding-evaluation workflow. Across the evaluation set, the system retrieved and reranked 3,000 evidence chunks and generated responses from the top-ranked evidence. Claim-level grounding evaluation identified 1,200 factual claims, of which 1,184 were judged to be supported by the retrieved evidence and 16 were classified as unsupported, resulting in an overall grounding accuracy of 98.7%. The results indicate that hybrid retrieval, reranking, conservative prompting, and claim-level grounding evaluation can support reliable evidence-grounded RAG responses when sufficient source evidence is available. The study also highlights the distinction between internal grounding accuracy and strict citation accuracy, motivating future work on explicit sentence-level citation generation and expanded human expert validation.

Cureus Journal of Computer Science.
University of North Texas (US), California State University, Fullerton (US)
Openalex Percentile: Top 13%
Topic Modeling
1.94
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon