Sentiment-oriented Transformer-based Variational Autoencoder Network for Live Video Commenting