We conduct an extensive experimental evaluation to verify the predictive performance of the domain-specific deep views and the deep co-training framework. The experimental results show that our method is markedly better than the existing deep learning and semi-supervised learning methods in the context of CQA systems.