Exploiting User Comments for Early Detection of Fake News Prior to Users' Commenting