Kernel-improved Support Vector Machine for semanteme data

摘要

The computation for the inner production of semanteme data and the Support Vector Machine (SVM) classification of semantic data are very difficult. In this paper, a novel kernel-based semanteme data classification method is proposed, and the (SVM) is extended to semantic SVM, called the Semanteme-based Support Vector Machine (SSVM). A new dissimilarity definition and a simple inner production computation method for semanteme data are presented, and the parameters for optimization selection in SSVM are also discussed. In the proposed algorithm, the solving support vector process based on the new inner production computation method remains a quadratic program problem but has a lower computation complexity. The SSVM is insensitive to outliers, and its classification capability for unbalanced data in actual datasets is analyzed. The experimental results demonstrate the average advantage of SSVM over algorithm C4.5, adaptive dissimilarity metric, and value difference metric in terms of classification and robust capability, indicating that the proposed method has promising performance.