Multiview Fuzzy Concept-Cognitive Learning With High-Order Information Fusion of Fuzzy Attributes
Jinbo Wang, Weihua Xu, Weiping Ding, Yuhua Qian
Abstract
Concept-cognitive learning (CCL) is an emerging computing paradigm that is widely employed in knowledge discovery. It considers concepts as the basic computing units, emphasizing the representation of knowledge through extent–intent pairs. Some studies explore CCL models on single view data, with a particular focus on fuzzy CCL models, demonstrating notable performance in classification tasks. However, data are always obtained from multiple views in reality, necessitating the crucial task of representing and integrating concepts across multiple views. Hence, this article proposes a novel multi-view fuzzy concept-cognitive learning (MVFCCL) model to address this issue. The process of multiview fuzzy concept cognition is first introduced to learn fuzzy concepts from each view. Specifically, the process provides an intraview fusion method to reconstruct fuzzy attributes by modeling both high-order information and correlation information, thereby enhancing the conceptual representation ability of fuzzy concepts. Then, the multiview fuzzy concept recognition process is established to predict new objects decision attributes by considering their similarity to the multiview fuzzy concept space. Finally, some experiments are conducted to examine the effectiveness of MVFCCL, including comparisons with other methods and ablation experiments for validating the contributions of each step. Experimental results show that the proposed MVFCCL can effectively represent and fuse knowledge from multiview data via fuzzy concepts.