Dynamic Multi-View Classification and Knowledge Fusion: A Fuzzy Concept-Cognitive Learning Perspective

Recently, multi-view data have grown significantly in practical scenarios. Compared with single-view data, they can comprehensively describe objects through diverse types of features. However, their inherent heterogeneity introduces new challenges for knowledge discovery, especially in dynamic environments. To effectively represent knowledge in dynamic multi-view data, this paper proposes a dynamic multi-view concept-cognitive learning (DMVCCL) model. First, a multi-view knowledge representation framework is established, which uses fuzzy three-way concepts as basic carriers. The natural hierarchical relationship between concepts is utilized to precisely represent knowledge in multi-view data. Then, for dynamic multi-view data, a clue-based dynamic concept updating mechanism is designed. This mechanism leverages the varying sensitivities of concepts at different granularity levels to data changes, enabling learning concepts at the optimal granularity level. Moreover, the weights of each view are assigned based on the representation capability of the learned concepts, and a multi-view classification method is designed using the similarity between concepts and data. Finally, a series of comparative experiments are conducted to verify the effectiveness of the proposed method.