Improving the Modality Representation with Multi-View Contrastive Learning for Multimodal Sentiment Analysis