Abstract: This paper deals with clustering for multi-view data. Multi-view clustering takes every data view into consideration, as more effective information can be used in the clustering process, its performance is better than that of the single-view clustering algorithm. However, most existing clustering algorithms consider every view equally in terms of importance, but if there is a poor quality view, it will seriously affect the clustering results. Due to differences in the quality of information that different views contain, the impacts on the clustering results are also different. This paper assigns different weights for different views according to their contribution rate on clustering, and proposes a new multi-view clustering algorithm — EWCMVC (entropy weighting centralized multi-view fuzzy clustering) by using the centralized strategy. The experimental results obtained both on synthetic and real datasets show that the proposed algorithm outperforms the most of existing multi-view clustering algorithms.

Key words: multi-view datasets, fuzzy clustering, entropy weighting, centralized strategy

摘要： 研究了多视角聚类问题，由于多视角聚类考虑到每个样本在多个视角的信息后进行聚类，并利用了更多的有效信息，因而较单视角聚类算法更优。目前绝大多数多视角聚类算法在聚类过程中认为各个视角同等重要，但是如果其中存在质量较差的视角，则会严重影响聚类的最终结果。不同的视角由于其包含信息质量的差异，对聚类最终结果的影响也是不同的。根据每个视角对聚类的贡献率赋予每个视角不同的权值，并利用中心化策略，提出了基于视角熵权重的中心化多视角模糊聚类（entropy weighting centralized multi-view fuzzy clustering，EWCMVC）算法。在人工数据集和实际数据集上的仿真结果验证了该算法聚类性能优于传统单视角和多视角聚类算法。

关键词: 多视角数据, 模糊聚类, 熵加权, 中心化策略