Research On Clustering Method With Robust Dual Concept Factorization

With the rapid development of information age,the effective acquisition of massive data becomes the key.Data mining,as an important field of rapid development in the information age,has been widely used in various industries.Cluster analysis is one of the earliest methods used in data mining.As an effective method to study data mining,cluster analysis finds out the potential relationship between data samples by classifying and dividing massive data.In clustering analysis,matrix decomposition can realize problem analysis of high-dimensional data into several low rank matrix problem analysis through data dimension reduction,Further,we can achieve effective clustering.In this paper,we mainly increase the duality and robustness of the function according to the relevant knowledge of concept decomposition.In view of the duality and robustness of the function in clustering analysis,we propose two corresponding algorithms for the decomposition of dual concepts,as follows:(1)A nonnegative matrix factorization algorithm(ODCF)based on orthogonal dual reconstruction is proposed.The purpose of this method is to remove the influence of non negative factors on the experimental results on the basis of non negative matrix,because in dealing with practical problems,the non negative of data is often difficult to guarantee,which also makes the research of the problem become limited.Therefore,considering the concept decomposition,from the original study of single clustering in sample aspect to the study of dual clustering in sample aspect and feature aspect,Then we do the dual concept decomposition algorithm on the basis of concept decomposition,and further optimize the model.We added two regular terms.Through the compression of data,the approximation of data is further improved.A large number of experiments show that the clustering effect of this method on benchmark data set is better than other similar algorithms.(2)A robust dual concept decomposition(DRCF)algorithm based on local sensitivity is proposed.Based on the dual concept decomposition,this method takes into account the influence of noise on the clustering results.In practical problems,Noise pollution has a greater impact on the clustering results,so we consider that the error loss is larger in the part with large noise,so we introduce the semi quadratic minimization loss function,which has better robustness in solving the large error part,Fully considering the effect of noise factor,the function is not sensitive to the influence of noise factor in a certain range,and the robustness of the function is good.A large number of experiments show that this method is better than some advanced algorithms in three clustering indexes.To sum up,this paper focuses on the duality and robustness of function in concept factorization.First of all,we consider whether we can construct concepts on the sample side and the feature side at the same time,Through this way,the data can be compressed,and the approximation degree and clustering effect of the data will be better,Furthermore,in order to reduce the influence of noise pollution on clustering results,we add robust graph regularization to the sample side and the special side,which makes our method more robust and clustering effect more significant.The two dual clustering methods proposed in this paper are of great research value both in academic research and practical application.