基于权值的局部保持半监督降维算法*

针对现有的半监督降维算法没有考虑存在于数据集中的大量未标记信息，不能得到最好的降维效果的问题。本文提出了一种改进的基于权值的局部保持半监督降维算法。该算法在保持正、负约束信息的同时，还利用距离权值来保持数据集所在的局部结构，从而提高降维效果。在UCI数据集上的实验表明，该算法能够提高降维的效果，尤其是在数据分布特性不满足流形结构时，仍能得到较好的聚类结果。

Weight local preserving based semi-supervised dimensionality reduction

The existing semi-supervised dimension reduction algorithms do not consider unlabeled information in the dataset, so they do not obtain the best result for dimension reduction. This paper presents an improved weight-based semi-supervised locality preserving dimensionality reduction algorithm. The algorithm not only maintains the positive and negative constraint information, but also keeps the local structure of dataset using the distance weights. Experiments on UCI datasets show that the algorithm can improve the effect of dimensionality reduction, and also get better clustering results in the data distribution structure which does not meet the manifold.