Abstract: Since the existing single view data competition clustering algorithm has poor performance on multiple views data, a view correlation factor based multi-view data competition clustering algorithm is proposed. Firstly, a view correlation factor is defined as the correlation between different views. Next, the view correlation factors are combined with spectral objective function maximum problem, and a joint objective function which can make full use of information from different views is built to make the information interaction between different views. By solving the joint objective function optimization problem, the optimized embedding matrices of each view are obtained. Then, the optimized embedding matrices are used in data competition clustering algorithm. The simulation results on synthetic and real life datasets show that the proposed algorithm can obtain better performance than existing data competition clustering algorithm.

Key words: clustering, data competition, aggregation field, multi-view, view correlation factor

摘要： 针对现有的单视图数据竞争聚类算法无法高效处理多视图数据的问题，提出了基于视图相关因子的多视图数据竞争聚类算法。首先，为了描述不同视图之间的相关性定义了一种视图相关性因子；然后，将视图相关因子与谱方法关于拉普拉斯矩阵的目标函数最大化问题结合，建立一个联合目标函数，使得不同视图之间的信息相互影响，以充分利用多视图的信息。通过解决联合目标函数的优化问题，得到每个视图的优化嵌入矩阵；最后，将得到的优化嵌入矩阵用于数据竞争聚类算法中。在人工和真实数据集上的仿真实验结果表明，新算法比现有的数据竞争聚类算法具有更高的聚类性能。

关键词: 聚类, 数据竞争, 聚合场, 多视图, 视图相关因子