Research On Spectral Clustering Algorithm Of High-dimensional Multi-view Data

As an excellent algorithm among clustering algorithms,Spectral clustering has made good progress in various fields in recent years.Compared with a single view,many real-world applications involve data collected from different views,and have high data dimensions and inevitable noise.Due to dimensional disasters,ineffective denoising,and poor results of multi-views integration,clustering on such high-dimensional and noisy datasets is still a challenge.Therefore,how to design an algorithm that can effectively process high-dimensional noisy data is of great significance for multi-view clustering.In the process of collecting data,the data is constantly increasing,and some data cannot be obtained all at once.Traditional spectral clustering clusters dynamically increasing datasets and requires repeated running of the algorithm,which leads to a longer time and lower efficiency.Therefore,how to design an algorithm so that the algorithm can effectively process incremental data while reducing time overhead is of great significance for multi-view clustering of incremental data.This thesis focuses on solving the clustering quality problems of two different types of data:high-dimensional noise and sample dynamic increase.The research contents are as follows:(1)For noisy high-dimensional multi-view data,in view of the advantages of l_2,1-norm in processing noisy data,this thesis proposes a robust self-weighted multi-view projection clustering algorithm based on l_2,1-norm(RSw MPC).The algorithm can simultaneously perform dimensionality reduction,noise suppression and learning local structure diagrams.At the same time,the obtained optimal graph can be directly used for clustering without further processing.Besides,the algorithm also uses a new method of automatically learning the optimal weight of each view,without generating additional parameters to adjust the weight.Extensive experiments have been conducted on different synthetic datasets and real datasets.The results show that compared with other latest related algorithms,the RSw MPC algorithm has great advantages in processing high-dimensional noisy data,and can obtain better clustering performance and robustness.(2)For incremental data,this thesis proposes a multi-view clustering algorithm for incremental data(IRSwMPC).The algorithm uses RSw MPC algorithm to cluster the existing data to obtain all clusters.The algorithm constructs a judgment index based on the eigenvalue increment.When the eigenvalue increment obtained after the newly added sample hypothesis belongs to each cluster is known,it can quickly determine the newly added sample category.There is no need to repeatedly run the clustering algorithm,which improves the algorithm's efficiency and reduces the time complexity.Experiments on real datasets verify the efficiency of the IRSwMPC algorithm in processing incremental data.