•  
  •  
 

Document Type

Article

Keywords

Robustness, Recommendation Systems, Clustering Methods, Adversarial Learning, Adversarial Attacks, Defense, Machine Learning

Abstract

Recommender systems (RSs) have become critical elements in modern instances of information and decision-support systems, resulting in a transformation of user experiences through highly personalized suggestions for an undeniably vast range of items. Although RSs have become commonplace, they continue to evolve, and their challenges, including sparsity, cold-start, scalability, and vulnerability to adversarial challenges remain. The use of clustering methods has proven highly effective in resolving these issues through the discovery and exploitation of latent user behaviour patterns, segmenting user groups that contribute more towards personalized and adaptable RSs. Additionally, adversarial learning has become a growing focus of study as a proposed solution for shielding and defending RSs, processes, and data from manipulation and attacks, resulting in greater resistance and trustworthiness. This study presents a systematic literature review (SLR) that explores the intersection of RSs, clustering methods, and adversarial learning. This paper synthesizes a critique of the latest hybrid recommendations, detailing motivations, challenges, directions for future study, and practical recommendations drawn from the examined studies. An SLR search of four academic databases, ScienceDirect (SD), IEEE Xplore (IEEE), Scopus, and Web of Science (WoS), delivered an initial yield of 843 studies; after filtration, 51 studies remained. All the retained articles were examined and characterized in terms of dataset details, techniques, frameworks, and performance. A significant gap in research has emerged regarding the overreliance on datasets from commercial and entertainment domains, with a notable scarcity of studies addressing critical domains such as healthcare, finance, and other critical fields where diverse data sources should invoke robust, secure, and trustworthy RSs recommendations. Future research is needed to develop adversarially robust RSs for high-stakes applications requiring stringent accuracy and safety standards. This review provides a rich critical examination of the literature to embolden the ideas and theories associated with clustering methods and adversarial learning working together within RSs. It offers concrete opportunities and directions for carrying out future work in developing next-generation secure, adaptive recommendation frameworks. These findings corroborate a change in perspective on designing systems that seek to develop an RSs that can withstand adversarial threats and promote the development of safer, fairer, and more reliable decision-support systems in a variety of domains.

Download

Share

COinS