期刊名称：International Journal of Advances in Soft Computing and Its Applications
印刷版ISSN：2074-8523
出版年度：2018
卷号：10
期号：3
出版社：International Center for Scientific Research and Studies
摘要：In the past, many dimensional reduction methods such as Local Linear Embedding (LLE) and Laplacian Eigenmaps (LE) have been successfully developed. However, in many real world applications, representing the dataset as un-directed graph, used in Laplacian Eigenmaps and Local Linear Embedding methods, is not complete. Approximating complex relationship as pairwise will lead to the loss of information. The natural way overcoming the information loss is to represent the dataset as the hypergraph. However, representing the dataset as the hypergraph will not lead to the perfection. The number of hyper-edges may be large; hence this will lead to high time complexity of the clustering methods or the classification methods when we try to apply the clustering/classification methods to this hypergraph dataset. Thus, in this paper, we develop the un- normalized hypergraph Laplacian Eigenmaps. Moreover, in the developed un-normalized hypergraph Laplacian Eigenmaps algorithm, we assume that the weights of all hyper-edges are equal to 1. This is not true at all in practical applications. Thus, in this paper, we will also develop the weighted un-normalized hypergraph Laplacian Eigenmaps and the weighted hypergraph semi-supervised learning method. Experimental results show that the weighted hyper- graph semi-supervised learning method achieves the highest accuracy performance measures.
