关键词: 数据挖掘, 多视图, 聚类, K-means, 样本权重

Abstract: In the era of big data, the ability of humans to collect, store, transmit and manage data has been increasingly improved. Various industries have accumulated a large amount of data resources, which are often multi-source and heterogeneous. How to effectively cluster these multi-source data (also known as multi-view clustering) has become one of the focuses of today’s machine learning research. The existing multi-view clustering algorithms mainly pay attention to the contribution of different views and features to the cluster structure from the “global” perspective, without considering the “local” information complementary differences between different samples. Therefore, this paper proposes a new sample-weighted multi-view clustering (SWMVC). The method weights each sample with different views and adopts alternating direction method of multipliers (ADMM) to learn sample weight, which can not only learn the “local” difference of weights among multiple views in different sample points, but also reflect the “global” difference of the contribution of different views to the cluster structure, and has better flexibility. Experiments on multiple datasets show that the SWMVC method has a better clustering effect on heterogeneous view data.

Key words: data mining, multi-view, cluster, K-means, sample weights