An Integrated PCA-DAEGCN Model for Movie Recommendation in the Social Internet of Things

Abstract

With the development of the Social Internet of Things (SIoT) and mobile technologies in recent years, movie recommendation systems have become popular in online movie recommendation that users may like to watch based on their historical movie viewing data monitored by the SIoT. This technology can bring considerable profits to online movie providers and has attracted the attention of a large number of related scholars. However, previous movie recommendation models based on autoencoders have insufficient model learning ability due to their defective features. Due to the errors in users’ operation, the user’s movie rating data will have some errors. The previous models cannot deal with this corrupted information, which leads to the degradation of their generalization performance. Presently, graph convolutional networks have made full progress in many fields, and they can outperform traditional methods. Therefore, in this work, we introduce a principal component analysis and denoising autoencoder integrated graph convolutional networks (PCA-DAEGCNs) for movie recommendation in the SIoT. The PCA-DAEGCN model uses the network structure of the graph autoencoder to obtain effective hidden features and, subsequently, uses denoising autoencoders to handle small changes in the feedback information. Finally, the captured hidden features of users and movies are used to derive the finally predicted scores. Comprehensive experiments show that the proposed PCA-DAEGCN is able to obtain far better efficiency than many comparative models.