Research On Fruit And Vegetable Recommendation Based On Graph Neural Network Considering Time Context

In the face of various e-commerce platforms,the number and variety of fruit and vegetable products,users often find it difficult to make a choice.By modeling the preference characteristics of users and commodities,the recommendation system recommends the commodities in line with their preferences for users,which alleviates the problem of information overload to a certain extent.However,the existing recommendation models of fruit and vegetable products only consider the interaction between users and fruit and vegetable products,but ignore the time context information in the interaction process.At the same time,the existing graph neural network recommendation model may have a smoothing problem after stacking multiple layers,which affects the recommendation effect.This dissertation proposes a fruit and vegetable product recommendation model based on graph neural network with time context.The model integrates the time context information into the graph attention mechanism to better learn the features of nodes.At the same time,the model divides the user-fruit and vegetable bipartite graph into subgraphs,and learns the higher-order features of nodes in the subgraphs,which alleviates the problem of over-smoothing of the graph neural network model after stacking multiple layers.The main research contents of this dissertation are as follows:(1)As traditional fruit and vegetable product recommendation models seldom consider time context information,the fruit and vegetable product recommendation model proposed in this dissertation adopts a time attenuation factor that can dynamically capture user interest changes over time,and combines it with the traditional graph attention mechanism to give neighbor nodes attention weight that integrates time context information.In the process of node information transmission,the contribution of neighbors can be distinguished and the influence of time context information can be taken into account.(2)In view of the problem that the traditional graph neural network recommendation model appears too smooth after stacking multiple layers,the fruit and vegetable product recommendation model proposed in this dissertation adopts an unsupervised subgraph partitioning method,which uses user characteristics and graph structure to identify users with similar preferences and divides similar users into the same subgraph.At the same time,fruit and vegetable products that users interact with directly are also divided into the same subgraph because they can reflect users’ preferences most directly,and high-order propagation is implemented within the subgraph to avoid noise interference and effectively alleviate the problem of over-smoothing.(3)Experiments were carried out using the real data crawled from Jingdong e-commerce platform,and the fruit and vegetable product recommendation model proposed in this dissertation was compared with other relevant recommendation models,which verified that the model proposed in this dissertation has certain advantages in fruit and vegetable product recommendation.Finally,combined with the recommendation model proposed in this dissertation,a prototype system of fruit and vegetable recommendation based on graph neural network is designed and implemented.