Document Classification based on Optimal Laprls
Ziqiang Wang, Xia Sun, Lijie Zhang, Qian Xu · Journal of Software · 2013
To effectively utilize a large number of unlabeled data and a small part of labeled data in the document classification problem, a novel semi-supervised learning algorithm called optimal Laplacian regularized least square (OLapRLS) is proposed in this paper. This algorithm first obtains the data-adaptive edge weights by solving the l 1 -norm optimization problem; then the normalized graph Laplacian is derived for revealing the intrinsic document manifold structure; finally, the Nystrom method-based low-rank approximation method is adopted to reduce the computational complexity in manipulating the large kernel matrix. Experimental results on three well-known document datasets demonstrate the effectiveness and efficiency of the proposed OLapRLS algorithm.