融合KL信息的多视图模糊聚类算法

现有多视图模糊C均值聚类（FCM）算法通常将一个多视图分解为多个单视图进行数据处理，导致视图数据聚类精度降低，从而影响全局数据划分结果。为实现高维数据和多视图数据的高效聚类，提出一种基于KL信息的多视图自加权模糊聚类算法。将多个视图信息及其权重进行拟合融入标准FCM算法，求解多个隶属度矩阵和质心矩阵。在此基础上，通过附加KL信息作为模糊正则项进一步修正共识隶属度矩阵并保持权重分布的平滑性，其中KL信息是视图隶属度与其共识隶属度的比值，最小化KL信息会使每个视图的隶属度偏向于共识隶属度以得到更好的聚类结果。实验结果表明，该算法相比于传统聚类算法具有更好的聚类效果和更快的收敛速度，尤其在3-Sources数据集上相比于MVASM算法的聚类精度、标准化互信息和纯度分别提升了7.46、15.34和5.48个百分点。

Multi-View Fuzzy Clustering Algorithm Fused with KL Information

Existing multi-view Fuzzy C-Means(FCM) clustering algorithms usually artificially decompose multi-view data into multiple single-view data for processing, reducing the clustering accuracy of view data and affecting the results of global data division.To achieve efficient clustering of high-dimensional and multi-view data, a multi-view self-weighted fuzzy clustering algorithm based on Kullback-Leibler(KL) information is proposed, fitting multiple view information and their weights into the standard FCM algorithm to solve multiple membership matrices and centroid matrices.On this basis, additional KL information is used as a fuzzy regular term to further correct the consensus membership matrix and maintain the smoothness of the weight distribution, where the KL information is the ratio of a view's membership to its consensus membership, and minimizing the KL information biases each view's membership towards consensus membership, resulting in improved clustering results.The results show that the proposed algorithm has an improved clustering effect and faster convergence speed than traditional clustering algorithms.In particular, the clustering Accuracy(ACC), Normalized Mutual Information(NMI), and Purity of the MVASM algorithm on the 3-Sources dataset increased by 7.46, 15.34, and 5.48 percentage points respectively.