Research On Data Visualization Algorithm Based On Hessian Regularization

Faced with an increasingly complex data of analysis, we should combine the advantages of human visual and subjective perception for analysis of the results of intuitive and interactive. That can help the users visualize from different perspectives a deep understanding of complex data, so data visualization is more intrigued for data mining and analysis researchers.In this paper,large and complex data visualization research frequently encountered in intransitive similarities and other problems, especially a study for complex data visualization in high concentration problems. We explore the complex data based on multiple map visualization algorithms. The main work is as follows:With the increase of the amount of complex data, the efficiency of data visualization is getting lower and lower, Based on Barnes-Hut algorithm of the advantages and disadvantages, this paper come up with a multiple map t-SNE algorithm which based on Barnes-Hut. Multiple map t-SNE algorithm time complexity is exponential, which limits the number of data visualization. Combining the accelerated algorithm,the time complexity is reduced to O(NlogN). In ensuring neighbor preservation rate of the algorithm, multiple maps t-SNE algorithm which makes visual efficiency of complex data has been significantly improved; For different test data sets by selecting a reasonable threshold to achieve, The results show that multiple map t-SNE algorithm to get a good run at the same speed ensures visual accuracy.Aiming at the high concentration of data in the visualization of complex data, the result of the visual effect is poor, this paper come up with multiple maps t-SNE algorithm based on Hessian regularization. Hessian regularization has better inference ability than Laplace regularization, and which has better maintain of the local manifold structure. Hessian regularization is introduced to multiple map t-SNE algorithm. It is better to prevent data after data visualization, so it can enhance the effect of algorithm visualization. The results show that, based on Hessian regularization multiple map t-SNE algorithm is a good solution to the similarity of delivery problems, prevent complex data visualization over fitting, the data highly concentrated problem also has good visual effects.