Analysis based on distributed representations of various parts images in four-scene comics story dataset