Retrieval-Augmented Generation for Large Language Models
Mapping the Evolution and Intellectual Structure of RAG Research
Global Research Trends in Retrieval-Augmented Generation
A Bibliometric Analysis of RAG for Large Language Models
Bibliometrics Artificial Intelligence Open Science
Status: Ongoing Research
Research Stream: Science of Research & Emerging AI
Methods: Bibliometric Analysis · Science Mapping · Computational Text Analysis
Project Overview
Retrieval-augmented generation (RAG) has emerged as an important approach for extending the capabilities of large language models by integrating information retrieval with generative artificial intelligence.
As research in this area expands rapidly, there is a need to systematically examine how the field has developed, who and what have shaped its intellectual structure, and which themes are emerging.
This project uses bibliometric analysis and science mapping to examine the evolution of research on retrieval-augmented generation for large language models.
The study is designed not only to describe publication growth, but also to investigate the intellectual, conceptual, and collaborative structures shaping the field.
Scope of the Study
The primary unit of analysis is:
Retrieval-Augmented Generation (RAG) for Large Language Models (LLMs)
This scope intentionally focuses on the modern RAG paradigm associated with large language models rather than broadly combining it with earlier retrieval-based generation approaches in traditional natural language processing.
Earlier retrieval-generation methods may provide important historical context, but they are conceptually distinguished from the primary unit of analysis.
The current study period focuses on research published through 2025, allowing the analysis to examine the emergence and rapid growth of the field while using a complete publication year as the endpoint.
Research Questions
The project is guided by the following broad research questions:
RQ1 — Publication Trends
How has scholarly research on retrieval-augmented generation for large language models evolved over time?
RQ2 — Research Influence
Who are the most influential authors, institutions, countries, publications, and sources contributing to RAG research?
RQ3 — Collaboration
What patterns of collaboration characterize the global RAG research community?
RQ4 — Intellectual Structure
What citation and knowledge structures define the intellectual foundations of RAG research?
RQ5 — Emerging Themes
What major themes, research clusters, and emerging topics characterize the development of RAG for large language models?
Search Strategy
A central methodological consideration in this project is how to identify relevant RAG research comprehensively.
Searching only for the exact phrase “retrieval-augmented generation” may exclude relevant studies that describe similar architectures, approaches, or mechanisms using different terminology.
The project therefore uses a broader metadata-based retrieval strategy.
Records are retrieved using relevant information available through:
- Titles
- Abstracts
- Keywords
- Concepts and topics
- Related metadata fields
The search incorporates a comprehensive set of terms associated with retrieval-augmented generation and large language models.
Retrieved records are subsequently screened for relevance according to predefined eligibility criteria.
This approach is designed to improve recall while reducing unnecessary selection bias introduced by relying on a single exact phrase.
Data Source
OpenAlex
The primary bibliographic data source for this project is OpenAlex.
OpenAlex provides open scholarly metadata that can support:
- Publication retrieval
- Author analysis
- Institutional analysis
- Country-level analysis
- Citation analysis
- Topic and concept analysis
- Collaboration networks
The use of an open bibliographic data source also aligns with the project’s broader commitment to reproducible and transparent research.
Research Workflow
The project follows a reproducible bibliometric workflow:
Define Scope → Develop Search Strategy → Retrieve Records → Screen for Relevance → Clean Metadata → Analyze → Visualize → Interpret → Reproduce
1. Define the Unit of Analysis
Clearly define RAG for LLMs as the primary conceptual domain.
2. Develop the Search Strategy
Construct a comprehensive metadata search strategy designed to balance precision and recall.
3. Retrieve Bibliographic Records
Retrieve scholarly metadata through OpenAlex.
4. Screen Records
Manually screen retrieved records using predefined inclusion and exclusion criteria.
5. Clean and Standardize Metadata
Address issues such as:
- Duplicate records
- Author name variants
- Institutional name variants
- Keyword inconsistencies
- Missing metadata
6. Conduct Bibliometric Analysis
Analyze publication growth, productivity, impact, collaboration, citation structures, and research themes.
7. Conduct Science Mapping
Develop network-based representations of the field.
8. Analyze Emerging Themes
Use keyword analysis and computational text methods to identify developing research topics.
9. Document the Workflow
Maintain transparent documentation of search, screening, cleaning, and analytical decisions.
Analytical Methods
The project combines descriptive bibliometrics, network analysis, science mapping, and computational text analysis.
Performance Analysis
Potential indicators include:
- Annual scientific production
- Citation patterns
- Most productive authors
- Most influential authors
- Leading institutions
- Leading countries
- Most influential publications
- Most relevant publication sources
Collaboration Analysis
Collaboration networks may examine:
- Author collaboration
- Institutional collaboration
- International collaboration
Intellectual Structure
Citation-based analyses may include:
- Co-citation analysis
- Bibliographic coupling
- Citation networks
Conceptual Structure
The conceptual development of the field may be examined through:
- Keyword co-occurrence
- Thematic mapping
- Thematic evolution
- Topic analysis
- Trend topics
Computational Text Analysis
In addition to traditional bibliometric methods, the project explores computational text analysis to identify meaningful terminology within the literature.
TF-IDF
Term Frequency–Inverse Document Frequency (TF-IDF) may be used to identify terms that are particularly informative within documents or groups of documents.
This can complement author-provided keywords and bibliometric concept metadata by identifying terminology directly from titles and abstracts.
The integration of bibliometric and text-analytic approaches may provide a richer understanding of how RAG research themes have evolved.
Tools
The reproducible workflow incorporates:
Data Retrieval
OpenAlex API · openalexR
Data Processing
R · Python
Bibliometric Analysis
Bibliometrix · Biblioshiny
Science Mapping
VOSviewer
Text Analysis
R · Python · TF-IDF
Version Control
Git · GitHub
Reproducible Reporting
Quarto
Reproducibility
A central objective of the project is to ensure that the research process can be understood and reproduced.
The project workflow is designed to document:
- Search terminology
- Search dates
- API queries
- Inclusion criteria
- Exclusion criteria
- Screening decisions
- Data cleaning procedures
- Analytical code
- Software versions
- Visualization parameters
Where appropriate, research materials will be shared through GitHub and open research infrastructure.
Methodological Contribution
Beyond the substantive findings about RAG research, this project also contributes to a broader methodological objective.
The workflow developed for this study will inform the development of a reproducible bibliometric research framework that can be adapted to other rapidly evolving research domains.
This creates a connection between two related projects:
RAG Bibliometric Study → Reproducible Bibliometric Workflow → Future Bibliometric Studies
Explore the Reproducible Bibliometric Workflow →
Expected Research Outputs
Potential outputs from this project include:
- Peer-reviewed journal article
- Conference presentation
- Reproducible GitHub repository
- Open research documentation
- Bibliometric datasets where licensing permits
- Analytical code
- Science maps and visualizations
- Methodological documentation
As outputs become publicly available, links will be added to this page.
Broader Research Significance
RAG represents an important development in the evolution of large language models because it connects generative systems with external sources of information.
Understanding the development of this research domain can provide insight into:
- How the field has evolved
- Which researchers and institutions are shaping its development
- How international collaboration is structured
- Which intellectual foundations influence current research
- Which topics are becoming more prominent
- Where future research opportunities may exist
The project also contributes to the broader INSIGHT Lab research agenda in emerging artificial intelligence and the science of research.
Project Connections
This project intersects with several areas of my broader research program:
Science of Research & Emerging AI
Understanding how rapidly developing AI research fields evolve.
Reproducible Research
Developing transparent workflows for bibliometric research.
Agentic AI
Examining emerging AI architectures and research trajectories.
Research Synthesis
Developing methods for systematically understanding accumulated scientific knowledge.