Neural Topic Modeling for Document Networks with Adjacent-Encoder