熵加权多视角协同划分模糊聚类算法

当前,基于协同学习机制的多视角聚类技术存在如下两点不足：第一,以往构造的用于各视角协同学习的逼近准则物理含义不明确且控制简单；第二,以往算法均默认各视角的重要性程度是相等的,缺少各视角重要性自适应调整的能力。针对上述不足：首先,基于具有良好物理解释性的Havrda-Charvat熵构造了一个全新的异视角空间划分逼近准则,该准则能有效地控制异视角间的空间划分相似程度；其次,基于香农熵理论提出了多视角自适应加权策略,可有效地控制各视角的重要性程度,提高算法的聚类性能；最后,基于FCM框架提出了熵加权多视角协同划分模糊聚类算法(entropy weight-collaborative partition-multi-view fuzzy clustering algorithm,简称EW-CoP-MVFCM)。在模拟数据集以及 UCI 数据集上的实验结果均显示,所提算法较之已有多视角聚类算法在应对多视角聚类任务时具有更好的适应性。

Collaborative Partition Multi-View Fuzzy Clustering Algorithm using Entropy Weighting

There are two weaknesses of current multi-view clustering technologies based on collaborative learning. Firstly, the approximation-criteria of collaborative learning between each view is not clear for its physical meaning and is too simple to control the approximation-performance. Secondly, the existing algorithms assume that the significance of each view is equal, which is obviously inappropriate from the viewpoint of adaptively adjusting the importance of each view. In order to overcome the above shortcomings, a novel approximation-criteria of cluster partition based on the Havrda-Charvat entropy is proposed to control the similarity of cluster partition between each view. Then, an adaptive weighting strategy for each view based on the theory of Shannon entropy is presented to control the significance of each view and enhance the performance of the clustering algorithm. Finally, the collaborative partition multi-view fuzzy clustering algorithm using entropy weighting (EW-CoP-MVFCM) is provided. As demonstrated by extensive experiments in simulation data and UCI benchmark dataset, the proposed new algorithm shows the better adaptability than the classical algorithms on the multi-view clustering problems.