Abstract: Multi-view clustering aims to cluster the data described by different feature sets. Most of the traditional algorithms directly cluster the original feature sets, and ignore the influence of some hidden information on the clustering performance. Some multi-view clustering methods have tried to find hidden information embedded in multi-view data and to cluster based on hidden information, but such algorithms will lose the original feature information to varying degrees. To solve these problems, a multi-view clustering algorithm that is integrated with sparse hidden view information learning is proposed. Firstly, in order to mine the potential sparse hidden view information, a sparse hidden view information learning model is proposed, and the sparse hidden view information is obtained by solving the model. Then, the collaborative learning between the original feature sets and the sparse hidden view information is implemented in the clustering process. Experimental results on real datasets show that the proposed algorithm outperforms the existing clustering algorithms.

Key words: multi-view clustering, sparse representation, hidden view information, collaborative learning

摘要： 多视角聚类的目的就是对由不同的特征集描述的数据进行聚类。传统算法大多直接对原始特征集聚类，而忽略了一些隐性信息对聚类性能的影响。已有一些多视角聚类方法试图发现嵌入在多视角数据中的隐性信息并基于隐性信息进行聚类，但此类算法会不同程度地损失原始特征的信息。针对此，提出了一种融合稀疏隐视角信息学习的多视角聚类算法。首先为了挖掘潜在的稀疏隐视角信息，提出了一种稀疏隐视角信息学习模型，通过求解该模型获得稀疏隐视角信息。然后在聚类过程中实现原始的特征集和稀疏隐视角信息的协同学习。在真实数据集上的实验结果表明，所提算法的聚类性能优于现有的聚类算法。

关键词: 多视角聚类, 稀疏表示, 隐视角信息, 协同学习