LLMs Integration in Recommender Systems: A Comprehensive Survey of Frameworks, Taxonomies and Applications

Shojaei, Ardeshir, Asadi, Shahla ORCID: https://orcid.org/0000-0002-8199-2122, Ardakani, Mostafa K. ORCID: https://orcid.org/0009-0002-5193-6987 and Safaei, Mahmood ORCID: https://orcid.org/0000-0002-3924-6927 (2026) LLMs Integration in Recommender Systems: A Comprehensive Survey of Frameworks, Taxonomies and Applications. IEEE Access, 14. 55943 -55968. doi:10.1109/ACCESS.2026.3681933

Preview

Text 16225 Shojaei, A et al. (2026) LLMs_Integration_in_Recommender_Systems_A_Comprehensive_Survey_of_Frameworks_Taxonomies_and_Applications.pdf - Published Version Available under License Creative Commons Attribution 4.0. Download (2MB) | Preview

Abstract

Recently, Large Language Models (LLMs) have proven to be powerful tools for tackling persistent challenges within recommendation systems. This systematic literature review (SLR) explores how LLMs are being integrated into recommender systems. By examining studies published between 2018 to 2025, this study aims to provide a comprehensive overview of existing approaches, real-world applications, and recent advances in this area. The study addresses three research questions including existing methodologies and frameworks for integrating LLMs into recommender systems, developing a comprehensive taxonomy of LLM-enhanced recommender systems, and examining domain-specific applications across a range of business contexts. Through our SLR method, we examined 55 primary studies and identified five major integration approaches: CLLM4Rec-based frameworks, knowledge graph–driven integration, LLMER-type architectures, the strategic placement of LLMs at various stages of the recommendation pipeline, and purpose-built frameworks such as LLM4Rec. To address the second research question, we introduce a taxonomy that captures cold-start mitigation approaches, integration strategies including embedding-based, prompt-based, fine-tuned, and agent-oriented methods, as well as system architectures and evaluation criteria. The third research question examines the practical impact of these systems across a range of real-world application domain. Recent studies indicate that recommendation systems powered by LLMs are reshaping sectors such as e-commerce, retail, news media, content platforms, and healthcare. In these settings, LLMs contribute to more accurate product recommendations, more personalized marketing initiatives, and stronger customer engagement. In healthcare contexts, these models assist practitioners by supporting patient data analysis and informing clinical decision-making processes. Our results indicate that recommender systems enhanced with LLMs achieve strong performance, while demonstrating an improved ability to capture semantic information and nuanced contextual relationships. This comprehensive review provides researchers and practitioners with structured insights into recent advances in recommender systems powered by LLMs.

Item Type:	Article
Article Type:	Article
Uncontrolled Keywords:	Large language models (LLMs); Cold-start problem; Recommender systems, Systematic literature review (SLR); Taxonomy; Framework
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science Q Science > QA Mathematics > QA75 Electronic computers. Computer science > QA76.575 Multimedia systems
Divisions:	Schools and Research Institutes > School of Business, Computing and Social Sciences
Depositing User:	Kamila Niekoraniec
Date Deposited:	08 May 2026 13:36
Last Modified:	08 May 2026 13:45
URI:	https://eprints.glos.ac.uk/id/eprint/16225

University Staff: Request a correction | Repository Editors: Update this record