Multi-view Proximity Learning for Clustering

In recent years, multi-view clustering has become a hot research topic due to the increasing amount of multi-view data. Among existing multi-view clustering methods, proximity-based method is a typical class and achieves much success. Usually, these methods need proximity matrices as inputs, which can be constructed by some nearest-neighbors-based approaches. However, in this way, neither the intra-view cluster structure nor the inter-view correlation is considered in constructing proximity matrices. To address this issue, we propose a novel method, named multi-view proximity learning. By introducing the idea of representative, our model can consider both the relations between data objects and the cluster structure within individual views. Besides, the spectral-embedding-based scheme is adopted for modeling the correlations across different views, i.e. the view consistency and complement properties. Extensive experiments on both synthetic and real-world datasets demonstrate the effectiveness of our method.