Abstract: In order to realize the personalized recommendation with high accuracy and security, a new framework of distributed semi-supervised recommendation system was proposed. Co-NCF model was established through the combination of the co-training of semi supervised learning method with deep collaborative filtering model based on deep learning. Consensus-based distributed gradient decent algorithm was employed to train the Co-NCF model, so as to build the distributed version of Co-NCF model. In the test of MovieLens dataset, the performance of this model was significantly better than that of the existing distributed NCF model.