Skip to Main Content
Data management support to enable effective and efficient information sharing in collaborative environments is critical, especially in semantics based search and retrieval. In this paper, a novel spectrum perturbation-based subspace classification is proposed to mine semantics and other useful information from a large-scale dataset by utilizing a lower-dimensional subspace to discriminate different classes of the dataset. Among the existing subspace-based approaches, the principal component (PC) subspace is the most prevailing one and has been well studied. After investigating previous work related to PC subspace, we found that none of them had considered the perturbation on spectrum when building the subspace learning models. However, such perturbation is of certain importance and is able to provide discriminant information that helps improve classification performance by measuring the closeness of each testing data instance towards a subspace model by a closeness score based on the spectrum perturbation. Each testing data instance is assigned to its closest class by searching the smallest closeness score. Experiments are conducted to evaluate our proposed subspace classifier using data sets from three different sources, and the experimental results show that it achieves promising results and outperforms comparative subspace classifiers as well as some other commonly used classifiers.
Date of Conference: 15-18 Oct. 2011