ニューラルネットを用いた文体のモデル化と情報圧縮の検討

抄録

In this paper, we assess models to distinguish the styles of writing in three different fields. The models are trained with the neural network and we employ the occurrence frequency of the end-of-sentence expressions in each document as feature to machine learning. For evaluating adequacy of each trained model to target problem, we compare the complexity of each trained model with AIC. Furthermore, showing the distribution of the examples on reduced variable space with the trained model, we confirmed its outcome as information compression from the original feature space.