【摘要】 与传统多视角聚类方法相比，多视角数据流聚类需要对持续到达的多视角数据进行实时融合与聚类分析，并要求聚类划分适应动态变化的数据分布。为实现此目标，本文提出了一种基于演化核嵌入的多视角数据流聚类算法(MSCEKE)。首先，通过设计多核融合机制，形成各视角聚类划分与共识核嵌入之间的交互协同，以提高多视角数据流的聚类性能。其次，结合支持向量域描述从历史数据中提取共识关键节点，利用共识关键节点的传播构建核嵌入的演化机制，从而适应多视角数据流动态变化的数据分布。最后，利用支持向量聚类从核嵌入中获取聚类划分。通过在多个真实数据上的测试，验证了本文所提方法的有效性。更多还原

【Abstract】 Compared to traditional multi-view clustering methods, multi-view data streaming clustering requires real-time integration and clustering of continuously arriving multi-view data, while adapting to dynamically changing data distributions.To achieve this goal, this paper proposes a multi-view data streaming clustering algorithm based on evolving kernel embedding(MSCEKE). First, by designing a multi-kernel fusion mechanism, the interaction and collaboration between clustering partitions of different views and consensus kernel embedding are formed to improve the clustering performance of multi-view data streams. Secondly, by combining support vector domain description to extract consensus key nodes from historical data, and forming the evolution of kernel embedding through the propagation of these key nodes, the algorithm adapts to the dynamically changing data distribution of multi-view data streams. Finally, support vector clustering is utilized to obtain the clustering partitions. Extensive experiments on several real world datasets validate the effectiveness of our proposed method.更多还原