The essence of recommendation methods for online shopping is to predict whether the user will buy specified products. The existing methods are usually modeled on a user–item-score matrix, which does not consider the transformation of users and items. The transformation is hidden in the behavior and reflected in the shopping sequence. In a smart city of the future, products can be expressed by sensors: for a given item, the changes in its audience can be observed from the sequence of users who have bought these products, and for a given user the changes in his or her need submitted as requirement can be obtained according to the sequence of purchased commodities. To better capture the shopping sequence and apply it to recommendations, we proposed an E-commerce recommendation algorithm similar to the recommendation problem in the smart city using a double-layer graph neural network (DLGNN) to model the user sequence and item sequence and predict the purchasing behavior of users. In the DLGNN, the sequence data, including the sequence of users who have bought a particular item and the sequence of purchased commodities from each user, are first modeled as graph-structure data. Then by using a graph neural network (GNN), one can express the state of each node of the graph as cells in the neural network. On this basis, the pretrained representation vectors of users and commodities are used to represent their global information. The state vectors and representation vectors can be used to model the purchasing behavior of users. Experiments based on a real dataset show that the proposed method yields better performance than existing recommendation algorithms.