A Three-Way Clustering Algorithm via Decomposing Similarity Matrices for Multi-view Data with Noise

摘要

The multiple views of data can provide complementary infor-mation to each other, a large number of studies have demonstrated that one can achieve the better clustering performance by integrating infor-mation from multiple views than using only a single view. However, iden-tifying the explicit cluster structure in the multi-view data with noise and reflecting uncertain relationships between objects and clusters is still a problem that has not been satisfactorily solved. To address the prob-lem, this paper propose a three-way clustering algorithm for multi-view data with noise. The algorithm is mainly divided into two stages. In the first stage, we decompose the similarity matrix of each view into the good data and the corruptions to elimínate the noise contained in the multi-view data. In the second stage, only the clean data of each view is used to obtain the consistency information, and the final three-way clustering results are generated based on the theory of three-way decisions. The experimental results show that the proposed algorithm has better clustering performance in dealing with multi-view data with noise.