Research On Multi-View Clustering Algorithm Based On Bipartite Graph Learning

With the rapid development of information technology,the way people obtain data has gradually become diversified from the previous simplification,that is,from single-view description to multi-view description.Therefore,multi-view learning has become a research hotspot in the fields of artificial intelligence and machine learning.Each object in multi-view data has rich information,and multi-view clustering divides multi-view data into different groups according to the similarity between data objects.As the amount of data increases,bipartite graph-based multi-view clustering has become an important research topic,which achieves efficient clustering by establishing a relationship matrix between data points and a small number of anchor points.In addition,due to the richness of multi-view data itself,most graph data are usually complex,including node attributes and structural relationships between different vertices.Most of the current methods use the first-order bipartite graph to learn the graph structure directly from the original multi-view features.However,the original data often has noise and redundant information,which greatly affects the accuracy of the clustering performance.Therefore,how to learn from Extracting effective feature and structure information from multi-view data is one of the problems to be solved in this paper.In addition,compared with single-view data,the data information between different views has consistency and diversity,but most multi-view clustering methods based on bipartite graphs focus on the learning of consistent information between views,ignoring the information between views.The diversity information,which is crucial to improve the clustering accuracy.How to utilize the consistency and diversity information between multi-view data at the same time becomes the second problem to be solved in this paper.Aiming at the above problems,this paper proposes a multi-view clustering method based on graph filtering and high-order bipartite graphs and a diversity-induced bipartite graph fusion multi-view clustering method.The main works are as follows:(1)A multi-view clustering method based on graph filtering and higher-order bipartite graphs is proposed,which considers the influence of noisy features and complex graph structure relations.Specifically,graph filtering is first performed on the original data feature space,and then a bipartite graph structure relationship between sample points and anchor points in each view is established by a two-step random walk method.Then,based on the constructed high-order bipartite graph,a multi-view graph fusion framework based on self-weighted bipartite graph is proposed,which can reduce the tedious selection of weight parameters,and finally obtain a joint bipartite graph.Experimental results on several benchmark datasets demonstrate that the proposed method achieves better clustering performance than existing multi-view clustering methods.(2)A diversity-induced bipartite graph fusion multi-view clustering algorithm(Di BGFMGC)is proposed,which considers both the consistency and diversity information among multiple views.Specifically,in order to remove the diverse part while maintaining the consistency of multiple views,the bipartite graph is first decomposed into a consistent part and a diversity part,where the diversity part is composed of intra-view constraints and inter-view inconsistency constraints,because This can better distinguish it from the original bipartite graph.After eliminating the diversity part of the original bipartite graph,the method utilizes the consensus part to finally obtain a consensus bipartite graph with a clear cluster structure.Experimental results show that the Di BGF-MGC method achieves better clustering results than existing methods on eight benchmark datasets.