Research On Muti-view Clustering Base On Graph Learning

With the continuous advancement of science and technology,a large amount of data has been generated in many scientific and industrial fields.These data are represented by a variety of characteristics,forming multi-view data.Therefore,multi-view learning to process this kind of data has gradually become a research hotspot in the fields of deep learning,artificial intelligence,neural networks,and big data.Multi-view clustering is one of the research directions in the field of multi-view learning.After years of research and development,although many achievements have been made and applied in real life,there are still some problems.For example,a multi-view clustering algorithm needs to construct a relationship graph in advance through multi-view data.The structure of multi-view data is complex and there are different degrees of noise,so the constructed relationship graph is susceptible to noise.Moreover,the multi-view clustering algorithm obtains a unified relationship diagram through weighted fusion,but the influence of the weight distribution on the clustering result is not fully considered in the fusion process.In addition,most multi-view clustering algorithms divide the construction and fusion of relational graphs into two independent steps,resulting in the loss of part of the multi-view data information.Therefore,based on the problems in the multi-view clustering algorithm mentioned above,this article mainly does the following work:Aiming at the problem that the existing multi-view clustering algorithm does not fully consider the influence of the weight distribution on the clustering results in the process of fusing the relationship graphs and the multi-view data has different degrees of noise,this paper designs graph-based self weighted multi-view clustering algorithm.Firstly,the relationship diagram of each view is obtained through adaptive neighborhood learning,and then each relationship diagram is self weighted and merged into a unified relationship diagram,and the view weight adjustment parameter is introduced into the model,and the noisy view is obtained by adjusting the weight distribution of the view.A smaller weight reduces the impact of noise on the clustering results,and the model can obtain a better optimal solution as the view weight changes.Finally,the final clustering results are obtained through rank constraint optimization in the unified relationship graph.Aiming at the problem that the existing multi-view clustering algorithm divides the construction and fusion of the relationship graph into two steps,which leads to the loss of part of the multi-view data information,this paper designs a graph based joint learning self weighted multi view clustering algorithm.The algorithm first calculates the relationship graphs of all views,and then generates a unified relationship graph through self weighted fusion.The unified relationship graph and the relationship graph are alternately optimized through the alternate direction multiplier method,and the unified relationship graph and the relationship graph are learned in a mutually reinforcing manner to make up for the information loss caused by the separation step.In addition,the view weight adjustment parameter is added to the unified relationship graph to reduce the influence of noise.Finally,the final clustering result is obtained through rank constraint optimization.In this paper,two multi-view clustering algorithms are designed,which are graphbased self weighted multi-view clustering algorithm and graph based joint learning self weighted multi-view clustering algorithm.Contrast experiments with related clustering algorithms on real world data sets to prove the effectiveness and superiority of the method proposed in this paper.