Research On Partition Based Multi-view Clustering Algorithm

In recent years,due to the variety of data acquisition methods,we often describe the same thing from the different angles,resulting in a large number of multi-view data.Most of these data have no corresponding label information,so it is very important to mine the valuable information from such unsupervised data.Clustering analysis is an important unsupervised learning method in machine learning and can effectively excavate valuable information from the unsupervised data.On the basis of understanding and analyzing predecessors' works,a series of multi-view clustering algorithms based on partition model are proposed,which are as follows:1)A multi-view co-training clustering algorithm based on structure information preservation is proposed.The algorithm obtains subspace clustering results using the dimensionality reduction techniques aiming to preserve the global and local structures in a single view,and then utilizes this clustering result to guide the subspace clustering results in other views.In this way,the proposed algorithm can obtain the cluster partition taking full account of multiple view data's within-class compactness,between-class separation and neighborhood relationship,which is helpful to improve the accuracy of clustering.Experiments on artificial data sets and real data sets prove the correctness and effectiveness of the algorithm.2)A multi-view clustering algorithm based on fuzzy partition is presented.Compared with the multi-view clustering algorithm based on hard partition,this algorithm can describe the fuzzy relationship between multi-view data and all clusters,and has better interpretability for the clustering results.In addition,this algorithm sets the different weights for different views.By optimizing the multi-view clustering's objective function,the optimal view weight can be obtained,thus reflecting the importance of different views.The experimental results show that this algorithm can not only obtain the fuzzy relationship between the multi-view data and the clusters,but also can optimize the effective view weight.3)A multi-view fuzzy clustering algorithm based on online learning strategy is proposed.The algorithm segments the multi-view data into the multiple blocks and designs an online updating model for the clustering center and the cluster membership.The data block is clustered in sequence to form the final clustering result.In this algorithm,when the multi-view fuzzy clustering is performed on the current data block,the synthetic clustering information from the previous data blocks must be considered,so as to effectively solve the memory limitation problem.Through experiments on artificial data sets and real datasets,it is proved that online multi-view fuzzy clustering algorithm realizes online model and ensures the accuracy of clustering.