Research And Application Of Multi-label Classification Algorithm Based On Relevance And Dimension Optimization

With the development of the era of big data,a large number of multi-type,lowdensity,high-dimensional and complex data have been generated.Among them,one kind of data has attracted wide attention--multi-label data.At present,many classification algorithms deal with single-label classification,but for multi-label sample data,multi-label classification algorithm must be used to solve the classification and sorting problems of multi-label sample data.In this paper,the correlation optimization strategy between tags,Laplacian matrix regularization strategy and loss function rationality strategy based on non-negative matrix decomposition are integrated.In this paper,Multi-label Learning Model Based on Laplacian Regularized Extreme Learning Machine(ML-Lap-RELM)and Multi-label Learning Model Based on Projection Gradient Non-negative Matrix Factorization(ML-PGNMF)are proposed,and a Music Recommendation System Based on Label Matrix is designed and implemented.The main research contents and achievements are as follows:The similarity matrix between samples is calculated by using the manifold regularization extreme learning machine.Combined with the Lap-ELM strategy to calculate the weight matrix,the Laplacian matrix was regularized,and the multi-label Learning Model Based on Laplacian Regularized Extreme Learning Machine(MLLap-RELM)was proposed.Six evaluation indexes of multi-label classification will be used in the experiment.Firstly,the parameter with the best classification effect will be found under different parameter conditions,and then compared with other multi-label classification algorithms to verify that it is better.The case analysis and experimental results show that THE ML-Lap-RELM algorithm is feasible and the data label partition effect is better.2.Aiming at the problem of poor convergence performance of NMF,this paper uses the alternate calculation results to make the final results as close to the real results as possible,and proposes a Multi-label Learning Model Based on Projection Gradient Non-negative Matrix Factorization(ML-PGNMF).The original matrix is decomposed by PGNMF and then the ML-Lap-RELM algorithm is used to classify the highdimensional data with multi-label to verify the effectiveness of the algorithm.Compared with other dimensionality reduction algorithms on high-dimensional multilabel data sets,it is verified that the ML-PGNMF algorithm is effective and feasible,and its decomposition matrix is more reliable and efficient.3.Developed the Music Recommendation System Based on Label Matrix,realized the user module,music player module,personalized recommendation module,administrator module and other functional modules,reasonably divided the algorithm operation,and effectively managed data.Research contributions: Through label relevance optimization,proposed the multi-label Learning Model Based on Laplacian Regularized Extreme Learning Machine(ML-Lap-RELM).Through feature information loss function,proposed the Multi-label Learning Model Based on Projection Gradient Non-negative Matrix Factorization(ML-PGNMF).Develop Music Recommendation System Based on Label Matrix.