Evidential Weighted Multi-view Clustering

Generally, the data to be clustered are from one single view. In real clustering applications, sometimes the data are insufficient so that it is difficult to learn an ideal cluster model. In such cases, multi-view data can be taken into consideration in the clustering task. However, the inconsistency cross views may increase the cluster uncertainty. In this research, a new clustering method for multi-view object data, called MvWECM (Multi-view Weighted Evidential C-Means) is introduced in the framework of belief functions. The proposed method can take consistency and diversity cross each view into account by incorporating the concept of view weights to measure the importance of each view. An objective function is defined to look for the best credal partitions over the different views. Experimental results on generated and UCI data sets show the advantage of the proposed method.