Collaborative filtering in dynamic networks based on deep auto-encoder

The collaborative filtering method in recommender systems can produce suggestions with good quality. However, this method suffers from the issues of cold start and data sparsity. To overcome these problems, other data sources must be used to identify users. As a source of information, social relationships between users can help solve these problems. Recommender systems, on the other hand, are set in environments where data are generated continuously and with high dimension over time. For example, users have new rates and new relationships. These are indicative of the dynamic nature of system data, the changes in which classical models are unable to manage in order to provide appropriate suggestions to users. Then a dynamic model is needed. In this research, a new hybrid dynamic recommender model, which utilizes deep auto-encoder networks, is described to close these gaps. In this model, users’ rating information and social relationships between them are used simultaneously to calculate similarity matrices between users at different timestamps. In each timestamp, the similarity matrix is given as input to the deep neural network, in which users are divided into different clusters. Users’ new behaviors over time will help update matrices values. The use of social relationships data and the consideration of users’ new behaviors over time solves the problems of classical methods, reduces recommending error, and increases user satisfaction. The proposed model is compared with state-of-the-art models; based on evaluation metrics. The results of execution on the real dataset show that the proposed model outperforms them.