Research On Dimensionality Reduction Algorithms Based On Neighbors Protection Embedding

With the rapid development of information technology, a variety of high-dimensional data emerge. Although these data provide detailed information to people, at the same time, they also bring a lot of redundant information which can only make difficult to the data processing. So, how to extract the really useful information from the huge amount of data become the attention focus and a major difficulty in the processing of the data.High dimension not only makes the data difficult to be intuitively understood, it also makes it difficult to be effective treatment by the existing methods like machine learning and data mining. Faced with a series of problems and difficulties brought by high-dimension, the data dimensionality reduction becomes one of the effective methods to address this problem.In this paper, we analyze the traditional classic data dimensionality reduction algorithm and propose an improved algorithm-embedding algorithm based on neighbor protection of the image resources and embedding algorithm based on neighbor protection of the text resources.The embedding algorithm based on neighbor protection of the image resources includes sub setting, distance transformation embedding, multidimensional scaling analysis. By doing experiments on a simulation platform, we can draw the conclusion that with appropriate parameter setting, this algorithm is superior to the classic dimensionality reduction algorithms in neighbor protection for image resources. The key technologies in the embedding algorithm based on neighbor protection of the text resources are the similarity value and the distance transformation embedding function. Besides, we build an experimental environment and finish the simulation experiment. By analyzing the experimental results, we can conclude that the algorithms based on neighbor protection embedding method have very good results on neighbor protective effect, and get a better performance than traditional dimensionality reduction algorithms.