Authorship identification based on support vector machine

In recent years, various techniques and features are used in research of authorship identification. We propose a method of authorship identification by learning features of each author based on support vector machine (SVM). SVM achieves high generalization performance even with input data of very high dimensional feature space. In addition, by introducing a kernel function, SVM can carry out a training ill high-dimensional spaces with a smaller computational cost independent of their dimensionality. We adopted I 1 features and had an experiment to identify 5 Japanese writers to confirm effectiveness of our method. As a result, we achieve the accuracy of more than 80%. In addition, we found that the frequency of the consecutive characters and kinds of characters are effective features in authorship identification.