Dyn-GCN: Graph Embedding via Dynamic Evolution and Graph Convolutional Network for Personal Recommendation

Graph convolutional network (GCN) and dynamic evolutionary model are the mainstream collaborative filtering technologies in recent years. Nevertheless, the initial feature vectors selection problem of the existing recommendation algorithms based on dynamic evolutionary models leads to unstable recommendation accuracy. In addition, the collaborative filtering method of GCN does not take into account the dynamic evolution law of graph networks. Based on this, this research adopts GCN to train the initial embedding of the dynamic evolution model to perform collaborative filtering recommendation. First of all, a heterogeneous graph network is constructed by applying explicit feedback information (rating scores) of users. Secondly, the embedding of users and items are propagated through the dynamic evolution model. Then, the final embedding is obtained by weighting the embedding of each layer, and the scores are predicted. Finally, according to the Adam optimizer, the initial embedding of the dynamic evolution model is trained in the form of mini-batch to minimize the loss function. Experimental results show that the proposed algorithm is superior to several compared excellent algorithms in recommendation performance.