Research on Shopping Recommendation Based on Graph Neural Network and Deep Learning

For anonymous or new users on shopping websites, when their historical interaction information is unknown, a model can be established based on the user's historical behavior sequence data to mine their long-term interests and short-term preferences, and predict the items that the user may click on in the future. Then, based on the shopping sequence, this paper recommends a collection of items to purchase next. Inspired by the graph neural network, this network is used to establish a connection model between shopping items. This method can capture the transformation relationship between items and obtain the correlation between nodes and their neighboring nodes in the conversation graph. Through this approach, noise interference from unrelated neighboring nodes can be reduced and recommendation performance can be effectively improved. In order to establish a short-term interest model for users, a fixed size mobile window was designed to integrate the user preferences of the last few consecutive shopping baskets. By using a recurrent neural network, we fuse the basket representation, global, and short-term user shopping preferences to obtain the final recommendation result. The experimental results on two public datasets show that the algorithm can effectively improve the recommendation accuracy of the next shopping basket.