Breaking down multi-view clustering: A comprehensive review of multi-view approaches for complex data structures

Multi-View Clustering (MVC) is an emerging research area aiming to cluster multiple views of the same data, which has recently drawn substantial attention. Various approaches have been proposed in the literature, from a unified objective function across views, transfer learning approaches, and early, intermediate, and late fusion strategies to combine data emanating from multiple views. Despite the increasing popularity of MVC methods, there is still a lack of systematic categorization for MVC methodologies. Existing literature reviews on MVC often overlook the importance of providing a cohesive classification, and deep learning has significantly impacted the field of MVC by effectively analyzing complex data structures. This research addresses the need for consolidation by classifying MVC approaches into two main types: generative and discriminative. The Discriminative approaches are further split into four subclasses, including a category specifically focused on deep learning-based methods. This study emphasizes the growing importance of deep learning in MVC approaches. By categorizing deep learning as a separate class of algorithms, we intend to emphasize its significant influence on the MVC. A systematic comparison among different classes of MVC algorithms on benchmark textual and image datasets was conducted to objectively assess the efficacy of different MVC methodologies. Therefore, this review comprehensively analyzed existing MVC paradigms and identified potential research directions.