基于超图正则化的概念分解及在数据表示中的应用

针对传统图模型的流形学习无法准确表达数据间多元几何结构信息的问题, 提出一种基于超图正则化的概念分解(HRCF) 算法. 该算法用一组具有相似属性的数据子集构建超边, 建立数据间高阶关系的超图模型. 通过在概念分解算法中增加超图正则项, 保持数据间多元几何流形结构, 提高了算法的鉴别性. 在Yale 库、USPS 库和TDT2 库上的实验表明, HRCF 算法明显提高了聚类的准确率和归一化互信息, 验证了算法的有效性.

Hyper-graph regularized concept factorization algorithm and its application to data representation

The manifold learning methods of the simple graph model ignored the high-order relationship between data points. Therefore, an algorithm, called hyper-graph regularized concept factorization(HRCF) is proposed. HRCF considers the high-order relationship of samples by constructing the hyper-edge in hyper-graph with a subset of data points sharing with some attribute. The concept factorization(CF) algorithm can preserve the high-order relationship of the manifold structure, by adding hyper-graph regulation term in clustering. Thus, the algorithm has more discrimination power. The experimental results on Yale, USPS and TDT2 database show that the proposed approach provides a better representation and achieves better clustering results in terms of accuracy and normalized mutual information, and verify the effectiveness of the proposed method.